Front cover SQL Performance Diagnosis on IBM DB2 Universal

Transcription

Front cover
SQL Performance
Diagnosis
on IBM DB2 Universal Database for iSeries
Discover the tools to identify SQL
performance problems
Unleash the capabilities of the SQL
Performance Monitors
Learn to query the Database
Monitor performance data
Hernando Bedoya
Elvis Budimlic
Morten Buur Rasmussen
Peggy Chidester
Fernando Echeveste
Birgitta Hauser
Kang Min Lee
Dave Squires
ibm.com/redbooks
International Technical Support Organization
SQL Performance Diagnosis on IBM DB2 Universal
Database for iSeries
May 2006
SG24-6654-00
Note: Before using this information and the product it supports, read the information in “Notices” on
page vii.
First Edition (May 2006)
This edition applies to Version 5, Release 3, Modification 0 of IBM i5/OS, product number 5722-SS1.
© Copyright International Business Machines Corporation 2006. All rights reserved.
Note to U.S. Government Users Restricted Rights -- Use, duplication or disclosure restricted by GSA ADP Schedule
Contract with IBM Corp.
Contents
Notices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Trademarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The team that wrote this redbook. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Become a published author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Comments welcome. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ix
ix
xi
xi
Part 1. Introduction to DB2 Universal Database and database performance tools . . . . . . . . . . . . . . . . 1
Chapter 1. Determining whether you have an SQL performance problem . . . . . . . . . .
1.1 Questions to ask yourself . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 How do you know that there is a problem? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Where is the problem occurring? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4 Did you ever have a good working situation? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5 Do SQL queries appear to have performance problems?. . . . . . . . . . . . . . . . . . . . . . . .
3
4
4
5
7
7
Chapter 2. DB2 Universal Database for iSeries performance basics . . . . . . . . . . . . . . . 9
2.1 Basics of indexing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.1 Binary radix tree indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.2 Encoded-vector index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Query engines: An overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Database architecture before V5R2M0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.2 Current database architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3 Query Dispatcher . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.4 Statistics Manager . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.5 SQE Optimizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.6 Data Access Primitives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.7 Access plan. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3 Star Join Schema . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.1 Queries in a Star Join Schema . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.2 Restrictions and considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.3 Lookahead Predicate Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Part 2. Gathering, analyzing, and querying database performance data . . . . . . . . . . . . . . . . . . . . . . . 29
Chapter 3. Overview of tools to analyze database performance . . . . . . . . . . . . . . . . .
3.1 Current SQL for a Job function in iSeries Navigator . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Print SQL information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3 Debug messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4 Index Advisor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5 Index Evaluator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6 The Database Performance Monitors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6.1 Detailed Monitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6.2 Summary Monitor or Memory Resident Database Monitor . . . . . . . . . . . . . . . . . .
3.7 Visual Explain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
31
32
34
36
39
40
43
44
46
48
Chapter 4. Gathering database SQL performance data. . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1 Types of SQL Performance Monitors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.2 Collecting monitor data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
© Copyright IBM Corp. 2006. All rights reserved.
iii
4.2.1 Starting a Detailed Database Monitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.2 Ending a Detailed Database Monitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.3 Enabling Database Monitors in ODBC clients. . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.4 Enabling Database Monitors in OLE DB clients . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.5 Enabling Database Monitors in JDBC clients . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.6 Enabling Database Monitors using an exit program . . . . . . . . . . . . . . . . . . . . . . .
4.3 Collecting monitor data using iSeries Navigator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1 Starting a Memory Resident or Summary Database Monitor . . . . . . . . . . . . . . . .
4.3.2 Starting a Detailed Database Monitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.3 Importing Database Monitors into iSeries Navigator. . . . . . . . . . . . . . . . . . . . . . .
4.4 SQL Performance Monitors properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4.1 Considerations for the SQL Performance Monitors in iSeries Navigator . . . . . . .
4.5 Summary or Detailed Database Monitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6 The Database Monitor record types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.1 Database Monitor record types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.2 The 1000 Record: SQL statement summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.3 The 3000 Record: Arrival sequence (table scan) . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.4 The 3001 Record: Using an existing index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.5 The 3002 Record: Temporary index created . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.6 The 3003 Record: Query sort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.7 The 3004 Record: Temporary file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.8 The 3006 Record: Access plan rebuild . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.9 The 3007 Record: Index evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.10 The 3010 Record: Host variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.11 The 3014 Record: General query optimization information. . . . . . . . . . . . . . . . .
4.6.12 The 3015 Record: SQE statistics advised. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.13 The 3019 Record: Rows retrieved detail. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.14 Record information for SQL statements involving joins . . . . . . . . . . . . . . . . . . .
4.6.15 New MQT record types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
52
54
54
57
58
59
59
59
65
69
71
75
75
77
77
80
82
83
85
86
87
87
88
89
90
90
90
91
92
Chapter 5. Analyzing database performance data using iSeries Navigator . . . . . . . . 93
5.1 Considerations before analyzing Database Monitor data . . . . . . . . . . . . . . . . . . . . . . . 94
5.1.1 Importing the Database Monitor data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.1.2 Reducing the analysis time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.2 Predefined database performance reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.2.1 Accessing the collected performance data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.2.2 SQL performance report information from Summary reports . . . . . . . . . . . . . . . 102
5.2.3 SQL performance report information from Extended Detailed reports . . . . . . . . 105
5.3 Modifying a predefined query report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.4 Query tuning example with SQL Performance Monitor reports. . . . . . . . . . . . . . . . . . 123
5.4.1 List Explainable Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
Chapter 6. Querying the performance data of the Database Monitor . . . . . . . . . . . .
6.1 Introduction to query analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2 Tips for analyzing the Database Monitor files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2.1 Using an SQL ALIAS for the Database Monitor table . . . . . . . . . . . . . . . . . . . . .
6.2.2 Using a subset of the Database Monitor table for faster analysis . . . . . . . . . . . .
6.2.3 Using SQL views for the Database Monitor table . . . . . . . . . . . . . . . . . . . . . . . .
6.2.4 Creating additional indexes over the Database Monitor table. . . . . . . . . . . . . . .
6.3 Database Monitor query examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.1 Finding SQL requests that are causing problems . . . . . . . . . . . . . . . . . . . . . . . .
6.3.2 Total time spent in SQL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.3 Individual SQL elapsed time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
iv
SQL Performance Diagnosis on IBM DB2 Universal Database for iSeries
133
134
135
135
135
136
136
137
138
139
140
6.3.4 Analyzing SQL operation types. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.5 Full open analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.6 Reusable ODPs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.7 Isolation level used . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.8 Table scan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.9 Temporary index analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.10 Index advised . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.11 Access plan rebuilt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.12 Query sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.13 SQE advised statistics analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.14 Rows with retrieved or fetched details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.15 Materialized query tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
141
142
147
147
148
151
155
159
163
167
170
175
Chapter 7. Using Collection Services data to identify jobs using system resources 179
7.1 Collection Services and Database Monitor data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
7.1.1 Starting Collection Services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
7.1.2 From iSeries Navigator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
7.1.3 Using Performance Management APIs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
7.1.4 V5R3 STRPFRCOL command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
7.2 Using Collection Services data to find jobs using CPU . . . . . . . . . . . . . . . . . . . . . . . . 183
7.2.1 Finding jobs using CPU with the Component Report . . . . . . . . . . . . . . . . . . . . . 183
7.2.2 Finding jobs using CPU with iSeries Navigator Graph History . . . . . . . . . . . . . . 189
7.2.3 Finding jobs using CPU with Management Central System Monitors. . . . . . . . . 191
7.3 Using Collection Services data to find jobs with high disk I/O counts . . . . . . . . . . . . . 193
Chapter 8. Analyzing database performance data with Visual Explain . . . . . . . . . . .
8.1 What is Visual Explain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.2 Finding Visual Explain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.3 Using Visual Explain with the SQL Script Center . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.3.1 The SQL Script Center . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.3.2 Explain Only . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.3.3 Run and Explain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.4 Navigating Visual Explain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.4.1 Menu options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.4.2 Action menu items . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.4.3 Controlling the diagram level of detail . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.4.4 Displaying the query environment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.4.5 Visual Explain query attributes and values . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.5 Using Visual Explain with Database Monitor data. . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.6 Using Visual Explain with imported data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.6.1 List Explainable Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.7 Non-SQL interface considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.8 The Visual Explain icons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
197
198
198
199
199
200
200
200
205
206
209
211
211
214
216
218
219
220
Part 3. Additional tips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Chapter 9. Tips to prevent further database performance problems . . . . . . . . . . . . .
9.1 Indexing strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1.1 Access methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1.2 Guidelines for a perfect index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1.3 Additional indexing tips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1.4 Index Advisor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2 Optimization of your SQL statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2.1 Avoid using logical files in your select statements . . . . . . . . . . . . . . . . . . . . . . .
229
230
230
230
232
233
238
238
Contents
v
9.2.2 Avoid using SELECT * in your select statements . . . . . . . . . . . . . . . . . . . . . . . .
9.2.3 Avoid using the relative record number to access your data . . . . . . . . . . . . . . .
9.2.4 Avoid numeric data type conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2.5 Avoid numeric expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2.6 Using the LIKE predicate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2.7 Avoid scalar functions in the WHERE clause . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3 Reorganizing your database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3.1 Index Evaluator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3.2 Improved reorganization support for tables and physical files . . . . . . . . . . . . . .
9.3.3 Fast Delete support. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
240
241
242
243
246
248
248
249
249
250
Appendix A. Tools to check a performance problem. . . . . . . . . . . . . . . . . . . . . . . . . .
WRKACTJOB command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
WRKSYSACT command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
WRKSYSSTS command. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
WRKOBJLCK command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
WRKJOB command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
iDoctor for iSeries Job Watcher . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
253
254
255
257
258
259
260
Related publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IBM Redbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Other publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Online resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
How to get IBM Redbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Help from IBM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
263
263
263
263
264
264
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
vi
Notices
This information was developed for products and services offered in the U.S.A.
IBM may not offer the products, services, or features discussed in this document in other countries. Consult
your local IBM representative for information on the products and services currently available in your area. Any
reference to an IBM product, program, or service is not intended to state or imply that only that IBM product,
program, or service may be used. Any functionally equivalent product, program, or service that does not
infringe any IBM intellectual property right may be used instead. However, it is the user's responsibility to
evaluate and verify the operation of any non-IBM product, program, or service.
IBM may have patents or pending patent applications covering subject matter described in this document. The
furnishing of this document does not give you any license to these patents. You can send license inquiries, in
writing, to:
IBM Director of Licensing, IBM Corporation, North Castle Drive Armonk, NY 10504-1785 U.S.A.
The following paragraph does not apply to the United Kingdom or any other country where such provisions are
inconsistent with local law: INTERNATIONAL BUSINESS MACHINES CORPORATION PROVIDES THIS
PUBLICATION "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS OR IMPLIED,
INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF NON-INFRINGEMENT,
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Some states do not allow disclaimer of
express or implied warranties in certain transactions, therefore, this statement may not apply to you.
This information could include technical inaccuracies or typographical errors. Changes are periodically made
to the information herein; these changes will be incorporated in new editions of the publication. IBM may make
improvements and/or changes in the product(s) and/or the program(s) described in this publication at any time
without notice.
Any references in this information to non-IBM Web sites are provided for convenience only and do not in any
manner serve as an endorsement of those Web sites. The materials at those Web sites are not part of the
materials for this IBM product and use of those Web sites is at your own risk.
IBM may use or distribute any of the information you supply in any way it believes appropriate without incurring
any obligation to you.
Information concerning non-IBM products was obtained from the suppliers of those products, their published
announcements or other publicly available sources. IBM has not tested those products and cannot confirm the
accuracy of performance, compatibility or any other claims related to non-IBM products. Questions on the
capabilities of non-IBM products should be addressed to the suppliers of those products.
This information contains examples of data and reports used in daily business operations. To illustrate them
as completely as possible, the examples include the names of individuals, companies, brands, and products.
All of these names are fictitious and any similarity to the names and addresses used by an actual business
enterprise is entirely coincidental.
COPYRIGHT LICENSE:
This information contains sample application programs in source language, which illustrates programming
techniques on various operating platforms. You may copy, modify, and distribute these sample programs in
any form without payment to IBM, for the purposes of developing, using, marketing or distributing application
programs conforming to the application programming interface for the operating platform for which the sample
programs are written. These examples have not been thoroughly tested under all conditions. IBM, therefore,
cannot guarantee or imply reliability, serviceability, or function of these programs. You may copy, modify, and
distribute these sample programs in any form without payment to IBM for the purposes of developing, using,
marketing, or distributing application programs conforming to IBM's application programming interfaces.
vii
Trademarks
The following terms are trademarks of the International Business Machines Corporation in the United States,
other countries, or both:
Eserver®
Eserver®
Redbooks (logo)
iSeries™
i5/OS®
AS/400®
™
DB2 Universal Database™
DB2®
DRDA®
First Run™
IBM®
OS/400®
Redbooks™
System i5™
System/38™
SQL/400®
The following terms are trademarks of other companies:
Java, JDBC, and all Java-based trademarks are trademarks of Sun Microsystems, Inc. in the United States,
other countries, or both.
Excel, Microsoft, Visual Basic, Windows, and the Windows logo are trademarks of Microsoft Corporation in
the United States, other countries, or both.
Other company, product, and service names may be trademarks or service marks of others.
viii
Preface
The goal of database performance tuning is to minimize the response time of your queries. It
is also to optimize your server’s resources by minimizing network traffic, disk I/O, and CPU
time.
This IBM® Redbook helps you to understand the basics of identifying and tuning the
performance of Structured Query Language (SQL) statements using IBM DB2® Universal
Database™ for iSeries™. DB2 Universal Database for iSeries provides a comprehensive set
of tools that help technical analysts tune SQL queries. The SQL Performance Monitors are
part of the set of tools that IBM i5/OS® provides for assisting in SQL performance analysis
since Version 3 Release 6. These monitors help to analyze database performance problems
after SQL requests are run.
This redbook also presents tips and techniques based on the SQL Performance Monitors and
other tools, such as Visual Explain. You’ll find this guidance helpful in gaining the most out of
both DB2 Universal Database for iSeries and query optimizer when using SQL.
Note: In this book, we use the name “SQL Performance Monitors” when using iSeries
Navigator. SQL Performance Monitor has two versions: Detailed Database Monitor and
Summary Monitor. We refer to the tool as “Database Monitors” when using a green screen
and accessing the tool by running the Start Database Monitor (STRDBMON) CL
command.
The team that wrote this redbook
This redbook was produced by a team of specialists from around the world working at the
International Technical Support Organization (ITSO), Rochester Center.
Hernando Bedoya is an IT Specialist at the IBM ITSO, in Rochester, Minnesota. He writes
extensively and teaches IBM classes worldwide in all areas of DB2 Universal Database for
iSeries. Before joining the ITSO more than five years ago, he worked for IBM Colombia as an
IBM AS/400® IT Specialist doing presales support for the Andean countries. He has 20 years
of experience in the computing field and has taught database classes in Colombian
universities. He holds a master degree in computer science from EAFIT, Colombia. His areas
of expertise are database technology, application development, and data warehousing.
Elvis Budimlic is Director of Development at Centerfield Technology, Inc. in Rochester,
Minnesota. His primary responsibility is development and support of Centerfield’s insure/SQL
toolset with a focus on DB2 Universal Database for iSeries SQL performance and
optimization. His area of expertise are database performance, software engineering, problem
diagnostics, and iSeries work management. Before joining Centerfield, he worked for IBM for
four years on the database SLIC development group and Mylex RAID software development
teams. He holds a bachelor degree in computer science from Winona State University in
Minnesota.
Morten Buur Rasmussen is a senior IT Specialist at the IBM Client Technology Center
(CTC), in La Gaude, France. He covers Europe and the Middle East in client facing areas of
IBM Eserver iSeries, database, and WebSphere performance. Before joining the CTC, he
worked for IBM Denmark and different European banks. He has 18 years of experience in the
ix
computing field. His areas of expertise are database technology, application development,
and iSeries work management.
Peggy Chidester is a Staff Software Engineer working on the database team in the IBM
Rochester Support Center. She has been a member of this team since being hired by IBM
seven years ago and specializes in query performance. She has also worked temporarily on
the performance team in the Support Center as well as a liason with development. Peggy has
given new employees an introduction to database and taught database support employees in
Milan, Italy, about database performance. She has written numerous documents for the
Software Knowledge Base.
Fernando Echeveste is a Staff Software Engineer for IBM in Rochester, Minnesota. He is a
member of the IBM DB2 Universal Database for iSeries performance team. He has 12 years
of experience in the computing field. His areas of expertise are database technology and
client-server architectures. Before joining IBM in Rochester more than seven years ago, he
worked for IBM Guadalajara as a software engineer doing development and support for Client
Access/400 components such as Remote SQL, Transfer Function, and Open Database
Connectivity (ODBC). He holds a bachelor of science degree in computer engineering from
Western Institute of Advanced Technological Studies in Guadalajara, Mexico.
Birgitta Hauser has been a Software Engineer since 1996, focusing on RPG and SQL
development on iSeries at Lunzer + Partner GmbH in Germany. She graduated with a degree
in business economics, and started programming on the AS/400 in 1992. She is responsible
for the complete RPG, ILE, and Database programming concepts for Lunzer + Partner’s
Warehouse Management Software Package. She also works in education as a trainer for
RPG and SQL developers. Since 2002, she has frequently spoken at the COMMON User
Group in Germany. In addition, she is chief editor of the iSeries Nation Network (iNN, the
German part of iSeries Nation), eNews, and the author of several papers focusing on RPG
and SQL.
Kang Min Lee is an IT Specialist at IBM, based in Korea. He has more than six years of
experience in working with the IBM System i platform. Prior to joining IBM Korea, he had four
years of experience in applications development using DB2 Universal Database for AS/400
and iSeries. His areas of expertise are database technology, database performance, and
application development. Kang Min currently supports database and applications on
Eserver i5 and iSeries in Korea and provides general System i technical support.
Dave Squires works in the iSeries Support Center database team in the United Kingdom
(UK), where he provides first and second level support to clients in the UK and the Middle
East. He has been working on the iSeries and System/38™ since he started working in the
computing field in 1984.
Thanks to the following people for their contributions to this project:
Thomas Gray
Marvin Kulas
Joanna Pohl-Miszczyk
Jenifer Servais
ITSO, Rochester Center
Mark Anderson
Robest Bestgen
Michael Cain
Kevin Chidester
Daniel Cruikshank
Jim Flanagan
Kent Milligan
x
Brian Muras
Denise Voy Tompkin
IBM Rochester
Peter Bradley
IBM UK
Become a published author
Join us for a two- to six-week residency program! Help write an IBM Redbook dealing with
specific products or solutions, while getting hands-on experience with leading-edge
technologies. You'll team with IBM technical professionals, Business Partners and/or
customers.
Your efforts will help increase product acceptance and customer satisfaction. As a bonus,
you'll develop a network of contacts in IBM development labs, and increase your productivity
and marketability.
Find out more about the residency program, browse the residency index, and apply online at:
ibm.com/redbooks/residencies.html
Comments welcome
Your comments are important to us!
We want our Redbooks™ to be as helpful as possible. Send us your comments about this or
other Redbooks in one of the following ways:
򐂰 Use the online Contact us review redbook form found at:
ibm.com/redbooks
򐂰 Send your comments in an e-mail to:
[email protected]
򐂰 Mail your comments to:
IBM Corporation, International Technical Support Organization
Dept. JLU Building 107-2
3605 Highway 52N
Rochester, Minnesota 55901-7829
Preface
xi
xii
Part 1
Part
1
Introduction to DB2
Universal Database and
database performance
tools
In this part, we introduce basic information about DB2 Universal Database for iSeries
performance. We also introduce the different tools for analyzing database performance.
This part includes the following chapters:
򐂰 Chapter 1, “Determining whether you have an SQL performance problem” on page 3
򐂰 Chapter 2, “DB2 Universal Database for iSeries performance basics” on page 9
Note: In this book, we use the name “SQL Performance Monitors” when using iSeries
Navigator. SQL Performance Monitor has two versions: Detailed Database Monitor and
Summary Monitor. We refer to the tool as “Database Monitors” when using a green screen
command.
1
2
1
Chapter 1.
Determining whether you have
an SQL performance problem
In this chapter, we explain how to determine if a performance problem is the result of poorly
performing SQL queries. We describe the various system tools and a methodology to help
you find an SQL performance problem.
This chapter guides you in:
򐂰 Asking the right questions to determine whether you have a performance problem
򐂰 Knowing what to check when you have a performance problem
򐂰 Determining if a performance problem is related to SQL
3
1.1 Questions to ask yourself
If you are reading this book, you most likely currently have an SQL performance problem,
have experienced an SQL performance problem in the past, or are interested in learning how
to diagnose SQL performance. An important step in looking into any SQL performance
problem is to ask a few questions to determine whether the problem is with your SQL or a
system-wide performance problem. To clarify the problem, you must ask yourself the following
questions:
򐂰 How do you know that there is a problem?
򐂰 Where is the problem occurring?
򐂰 Did you ever have a good working situation?
– If you have a job or program that used to run well but doesn’t do so now, what changed
since it last ran well?
򐂰 Do SQL queries appear to have a performance problem?
– If the problem is with multiple queries, is there any commonality among the queries?
– If the problem is with one query, what is the change in run time?
– Have you ever tuned your queries for performance?
We explain each of these questions in more details in the sections that follow.
1.2 How do you know that there is a problem?
This question is very basic, yet vital. Identifying how you know whether there is a problem
determines the steps to take to analyze the problem.
For example, you are reading a report that shows that CPU spiked during a time frame. Is this
a problem? Is CPU spiking bad? Are you seeing a trend or is this a one-time occurrence?
During the time frame, did you receive complaints from users saying that they were unable to
get their work done? Such complaints would be a strong indicator that there is an actual
problem.
We recommend that you create a spreadsheet to document and keep a history of problems.
Table 1-1 shows an example of how problem record keeping can be helpful in problem
solving.
Table 1-1 Example of problem record keeping
Date of
problem
Time of
problem
User
reporting
Problem
job
Reported
problem
How problem
discovered
How problem
resolved
Data
gathered
01/05/05
13:01:00
Sales
Report
QZDASOINIT
Report
taking longer
to generate
User found
Note 1
SQL
Performance
Monitor
started
Note 1: The queries to generate the sales report usually run in a dedicated memory pool. However, for some reason still
unknown, the pool identifier in the QZDASOINIT prestart job was changed so that the queries started to run in *BASE
along with some other high demanding jobs on the system. By changing the prestart job entry, the queries went back to
run in their dedicated memory pool and the response time reverted to what it was in the past.
In the example shown in Table 1-1, it could look like an SQL problem since an SQL query was
taking a long time. However, further analysis showed that other work done in the pool was the
cause of the problem. In Appendix A, “Tools to check a performance problem” on page 253,
4
we explain how to use the system commands and tools to investigate whether the problem is
related to the system or to an SQL query. There are a number of solutions to try to resolve the
problem shown in Table 1-1:
򐂰 Separate the QZDASOINIT jobs into their own pool.
For more information, see Software Knowledge Base document “Separating Batch Work
from *BASE”, document number 349033974 on the Web at:
http://www-912.ibm.com/s_dir/slkbase.NSF/1ac66549a21402188625680b0002037e/9fa68bd7573e48
af862565c2007d3d9b?OpenDocument
򐂰 Verify that the QPFRADJ system value is set to automatically adjust memory in the pools.
򐂰 Add more memory to the pool.
1.3 Where is the problem occurring?
Another important step in analyzing an SQL performance problem is to identify where the
problem is occurring. It is helpful to understand the components of work involved whenever a
user executes any SQL request. Figure 1-1 illustrates the different components of work
involved in the execution of an SQL request.
END
Output Results
User Display I/O
Communications
RunTime
•
•
•
•
•
•
•
Open
Processing
• ODP creation
• Database authentication
Optimization
• Access plan creation
• Index estimates
Data
Processing
Database
Request
Authentication
User Display I/O
Journaling
Index maintenance
Constraint enforcement
Locking
Trigger processing
Disk operations
CPU operations
Process Request
BEGIN
Figure 1-1 Components of work
Isolate the problem to the smallest number of factors as possible. Before we start analyzing
the SQL or database request, we must understand that other variables are involved in the
total response time such as:
򐂰
򐂰
򐂰
򐂰
User display I/O
Authentication
Data processing
Communications
Chapter 1. Determining whether you have an SQL performance problem
5
You might ask some additional questions to find where the problem is occurring, such as:
򐂰 Is the problem occurring only in one pool?
– What activity is occurring in the pool?
򐂰 Are you having a performance problem running tasks on the iSeries server, even simple
commands?
򐂰 Is the problem occurring only within specific jobs, that is batch or QZDASOINIT?
򐂰 Do you hear only from one set of users when the problem occurs?
– What is the commonality among the users?
򐂰 Does the problem occur only when going through a specific interface, such as
WebSphere, SQL, or Query/400?
򐂰 Are the jobs having problems running remotely or locally?
򐂰 Can the problem be isolated to a specific program?
– Is SQL embedded?
– What type of program is it?
򐂰 Can the problem be isolated to a specific SQL statement?
You need to examine all of the answers to these questions to see if you can isolate the
problem to jobs running SQL. If the jobs that have a problem are QZDASOINIT or QSQSRVR,
then it is likely that they are running SQL. QRWTSRVR jobs are quite often used to run SQL,
but are also used if you are using distributed data management (DDM) files. When the
problem occurs only in certain jobs or for a certain group of users, you must review the
environment, such as communications, pools, job priorities, and so on.
After you isolate the problem and are now certain that the problem is on the SQL request, you
must understand what it takes to execute an SQL request. Upon the execution of an SQL
request, three main operations are done as shown in Figure 1-2:
򐂰 Optimization time
򐂰 Open processing time
򐂰 Run time
Output Results
RunTime
•
•
•
•
•
•
•
Journaling
Index maintenance
Constraint enforcement
Locking
Trigger processing
Disk operations
CPU operations
Open
Processing
• ODP creation
• Database authentication
Optimization
• Access plan creation
• Index estimates
Process Request
Figure 1-2 Components of work: Database request
6
Identify and minimize
bottlenecks in these
areas
In Figure 1-2, you can see which operations affect optimization time, open processing time,
and run time. It is key that you identify and minimize the bottlenecks in these three main areas
of the processing of an SQL request.
Understanding the tools that we describe in this book will help you to identify the bottlenecks
in these areas.
1.4 Did you ever have a good working situation?
A good working situation is when the system or an application is running without any
performance problems. Knowing if you had a good working situation involves understanding
the history of an application or the system. Some SQL performance problems are caused by:
򐂰
򐂰
򐂰
򐂰
The introduction of a new application
The application of new program temporary fixes (PTFs)
An upgrade to a newer i5/OS release
Changes to system values
In this book, you learn how to use some tools to gather performance information prior to any
major change on the system. You learn how to determine SQL performance problems in
cases where previously there was a good working situation or where you are unsure whether
you previously had a good working situation. In a scenario where you know that you had a
good working condition, it is vital to document the timeline of what happened since the system
or application last ran well. Make sure that you document any changes, such as those that we
previously listed.
1.5 Do SQL queries appear to have performance problems?
It might be difficult to answer this question if just one SQL query has a problem or if multiple
SQL queries have a problem, since the issue is often found at the job level. In the following
chapters, we show you how to use a different set of tools to investigate which queries are
having problems, if you have not already made that determination. It is important to
differentiate between one query having a performance problem and many SQL queries
having a problem.
򐂰 One SQL query having a performance problem
When it appears that a single SQL query has a performance problem, you must know the
run time of the specific query before the performance problem appeared. Additionally, you
must know how you gather the runtime data.
򐂰 Multiple SQL queries having performance problems
In situations where it appears that multiple SQL queries have performance problems, you
must ask additional questions to try to find any commonality among the queries:
– Do the queries that have the performance problem all use the same table?
– Does the problem appear to be with a specific type of query, such as left outer joins or
updates?
– Do only specific users run the queries?
– Do all the queries run in the same pool?
– Have you ever tuned your queries for performance?
When multiple queries have a problem, it indicates the need for you to review the
environment and examine such aspects as communications, pools, job priorities, and so
on. See Appendix A, “Tools to check a performance problem” on page 253, for more
information.
Chapter 1. Determining whether you have an SQL performance problem
7
One question to keep in mind when you are examining SQL performance problems is: “Did
you ever tune your queries?” SQL queries require the existence of indexes to obtain statistics
and for implementation (access of data) of the query request. For more information about an
adequate indexing strategy, refer to the white paper Indexing and statistics strategy for DB2
UDB for iSeries
http://www.ibm.com/servers/enable/site/education/abstracts/indxng_abs.html
In the following chapters, we explain how to use SQL Performance Monitors and other tools to
determine if your queries need to be tuned for performance. In Chapter 9, “Tips to prevent
further database performance problems” on page 229, we provide tips to help you avoid SQL
performance problems.
8
2
Chapter 2.
DB2 Universal Database for
iSeries performance basics
In this chapter, we introduce some of the basic concepts of SQL performance on DB2
Universal Database for iSeries. We discuss the indexing technology on DB2 Universal
Database for iSeries. We also introduce the query engines on DB2 Universal Database for
iSeries, Classic Query Engine (CQE) and SQL Query Engine (SQE).
In addition, we discuss Star Join Schema support on DB2 Universal Database for iSeries with
a key feature called Lookahead Predicate Generation.
9
2.1 Basics of indexing
DB2 Universal Database for iSeries has two kinds of persistent indexes:
򐂰 Binary radix tree indexes, which have been available since 1988
򐂰 Encoded-vector indexes (EVIs), which became available in 1998 with V4R3
Both types of indexes are useful in improving performance for certain kinds of queries. In this
section, we introduce this indexing technology and how it can help you in SQL performance.
2.1.1 Binary radix tree indexes
A radix index is a multilevel, hybrid tree structure that allows a large number of key values to
be stored efficiently while minimizing access times. A key compression algorithm assists in
this process. The lowest level of the tree contains the leaf nodes, which house the address of
the rows in the base table that are associated with the key value. The key value is used to
quickly navigate to the leaf node with a few simple binary search tests.
Figure 2-1 shows the structure of a binary radix tree index.
ROOT
Test
Node
MISS
ISSIPPI
002
AR
OURI
003
Database Table
IZONA
005
KANSAS
001
IOWA
004
001
ARKANSAS
002
MISSISSIPPI
003
MISSOURI
004
IOWA
005
ARIZONA
...
...
Figure 2-1 Binary radix tree index
Thus, a single key value can be accessed quickly with a small number of tests. This quick
access is pretty consistent across all key values in the index, since the server keeps the depth
of the index shallow and the index pages spread across multiple disk units.
The binary radix tree structure is good for finding a small number of rows because it can find
a given row with a minimal amount of processing. For example, using a binary radix index
over a customer number column for a typical online transaction processing (OLTP) request,
such as “find the outstanding orders for a single customer,” results in fast performance. An
index created over the customer number field is considered the perfect index for this type of
10
query because it allows the database to focus on the rows that it needs and perform a
minimal number of I/Os.
2.1.2 Encoded-vector index
To understand EVIs, you should have a basic knowledge of bitmap indexing. DB2 Universal
Database for iSeries does not create permanent bitmaps. SQL creates dynamic bitmaps
temporarily for query optimization.
The need for newer index technologies has spawned the generation of a variety of similar
solutions that can be collectively referred to as bitmap indexes. A bitmap index is an array of
distinct values. For each value, the index stores a bitmap, where each bit represents a row in
the table. If the bit is set on, then that row contains the specific key value.
Table 2-1 shows the bitmap representation of an index.
Table 2-1 Bitmap index
Key
Bit-Array
Arkansas
10000000110000010110
Arizona
01000100000011010000
...
Virginia
00000011000000100000
Wyoming
00001000000100000001
With this indexing scheme, bitmaps can be combined dynamically using Boolean arithmetic
(ANDing and ORing) to identify only those rows that are required by the query. Unfortunately,
this improved access comes with a price. In a very large database (VLDB) environment,
bitmap indexes can grow to an ungainly size. For example, in a one billion row table, you
might have one billion bits for each distinct value. If the table contains many distinct values,
the bitmap index quickly becomes enormous. Usually, relational database management
systems (RDBMSs) rely on a type of compression algorithm to help alleviate this growth
problem.
An EVI is created using the SQL command CREATE ENCODED VECTOR INDEX as in the
following example:
CREATE ENCODED VECTOR INDEX MySchema.EVI_Name
ON MySchema.Table_Name (MyColumn)
WITH n DISTINCT VALUES
An EVI is an index object that is used by the query optimizer and database engine to provide
fast data access in decision support and query reporting environments. EVIs are a
complementary alternative to existing index objects (binary radix tree structure, logical file, or
SQL index) and are a variation of bitmap indexing. Because of their compact size and relative
simplicity, EVIs provide faster scans of a table that can also be processed in parallel.
An EVI is a data structure that is stored basically as two components:
򐂰 Symbol table
The symbol table contains a distinct key list, along with statistical and descriptive
information about each distinct key value in the index. This table maps each distinct value
to a unique code. The mapping of any distinct key value to a 1-, 2-, or 4-byte code provides
a type of key compression. Any key value, of any length, can be represented by a small
bytecode. Additional key information, such as first and last row and number of
occurrences, helps to get faster access to the data.
Chapter 2. DB2 Universal Database for iSeries performance basics
11
򐂰 Vector
The vector contains a bytecode value for each row in the table. This bytecode represents
the actual key value found in the symbol table and the respective row in the database
table. The bytecodes are in the same ordinal position in the vector, as the row it represents
in the table. The vector does not contain any pointer or explicit references to the data in
the table.
Figure 2-2 shows the components of an EVI.
Vector
Row
Number
Symbol Table
Key
Value
Code
First
Row
Last
Row
Count
Arizona
1
1
80005
5000
Arkansas
2
5
99760
7300
Virginia
37
1222
30111
340
Wyoming
38
7
83000
2760
...
Code
1
1
2
17
3
18
4
9
5
2
6
7
7
38
8
38
9
1
Figure 2-2 Encoded-vector index
Note: Because the vector represents a relative record number list, an EVI cannot be used
to order records. EVIs also have a limited use in joins.
When executing queries that contain joins, grouping, and ordering, a combination of binary
radix indexes and EVIs might be used to implement the query. When the selected row set is
relatively small, a binary radix index usually performs faster access. When the selected row
set is roughly between 20% and 70% of the table being queried, table probe access using a
bitmap, created from an EVI or binary radix index, is the best choice.
Also, the optimizer and database engine have the ability to use more than one index to help
with selecting the data. This technique might be used when the local selection contains AND
or OR conditions, a single index does not contain all the proper key columns, or a single index
cannot meet all of the conditions. Single key EVIs can help in this scenario since the bitmaps
or relative record number (RRN) lists created from the EVIs can be combined to narrow down
the selection process.
Recommendation for EVI use
EVIs are a powerful tool for providing fast data access in decision support and query reporting
environments. However, to ensure the effective use of EVIs, you must implement them using
the guidelines:
Create EVIs on:
򐂰 Read-only tables or tables with a minimum of INSERT, UPDATE, and DELETE activity
򐂰 Key columns that are used in the WHERE clause: local selection predicates of SQL
requests, and fact table join columns when using Star Join Schema support
12
򐂰 Single-key columns that have a relatively small set of distinct values
򐂰 Multiple-key columns that result in a relatively small set of distinct values
򐂰 Key columns that have a static or relatively static set of distinct values
򐂰 Nonunique key columns, with many duplicates
Create EVIs with the maximum bytecode size expected:
򐂰 Use the WITH n DISTINCT VALUES clause on the CREATE ENCODED VECTOR INDEX
statement.
򐂰 If unsure, consider using a number greater than 65535 to create a 4-byte code, avoiding
the EVI maintenance overhead of switching bytecode sizes as additional new distinct key
values are inserted.
When loading data, keep in mind that:
򐂰 You drop EVIs, load the data, and then create EVIs.
򐂰 EVI bytecode size is assigned automatically based on the number of actual distinct key
values found in the table.
򐂰 The symbol table contains all key values, in order; there are no keys in the overflow area.
2.2 Query engines: An overview
Data is the key. Quick and reliable access to business data is critical to making crucial
business decisions. A robust database management system (DBMS) has excellent
performance capabilities and automated, built-in management and administration
functionality. It allows businesses to concentrate on making decisions based on the
information contained in their database, rather than managing the database.
Integrated into IBM OS/400® (i5/OS), DB2 Universal Database for iSeries has its roots in the
integrated relational database of the IBM System/38, the predecessor of the AS/400 and
iSeries servers. Although the database was always relational in nature, native file operations
were used to access the data.
With the debut of the AS/400 in 1988 came the introduction of SQL on the platform. SQL is an
industry standard (SQL 2003) to define database objects (Data Definition Language (DDL))
and manipulate database data (Data Manipulation Language (DML)). SQL provides an
alternative and additional method for accessing data. Both SQL and native methods can
coexist. Focusing primarily on OLTP applications, the database has satisfied customer
requirements for well over 20 years.
More recently a new breed of applications started to dominate development efforts. These
applications are designed to accommodate rapidly changing business needs and processes.
To address the issues and satisfy the demands of the new application world, IBM considered
the following options:
򐂰 Continue to enhance the existing product
򐂰 Acquire a new database technology
򐂰 Re-engineer the existing product
The continual enhancement of the product did not seem to be a viable proposition. The
increasing development resources required to maintain the existing code resulted in a
reduction of resources available to provide new functionality in a timely manner.
13
Acquiring a new database technology would compromise the basic tenets that distinguish the
iSeries from the rest of the industry. These include the integration of the database within
OS/400 and the ease-of-use characteristics of the database that minimize administration
efforts. Losing these characteristics would significantly reduce the cost of ownership benefits
of the iSeries.
Re-engineering the existing product was a more practical solution. However, this could easily
become an overwhelming and potentially unsustainable task if an attempt was made to
re-engineer the entire product. It could also impact portions of the product that continue to
provide solid and efficient support to existing applications and functions.
After considering the options, IBM chose to re-engineer the product. We did so with the
added decision to focus only on those aspects of the product for which re-engineering offered
the greatest potential. The potential offered the ability to:
򐂰 Support modern application, database, and transactional needs
򐂰 Allow the continued development of database functionality in an efficient and timely
manner
򐂰 Maintain and enhance the self-managing value proposition of DB2 Universal Database for
iSeries
򐂰 Provide a foundation to handle increasingly more complex query environments
򐂰 Improve query performance consistency and predictability
򐂰 Incorporate state-of-the-art techniques
In line with this decision, the query engine was identified as an area that would benefit
substantially from such a re-engineering effort. The best current technologies and algorithms,
coupled with modern object-oriented design concepts and object-oriented programming
implementation, were applied in the redesign of the query engine and its components.
To guarantee existing applications continue to work and to make new or even existing
applications profit from the new designed product, IBM decided to implement an additional
query engine. The newly redesigned query engine in DB2 Universal Database for iSeries is
the SQL Query Engine. The existing query engine is referred to as the Classic Query Engine.
Both query engines coexist in the same system.
The staged implementation of SQE enabled a limited set of queries to be routed to SQE in
V5R2. In general, read-only single table queries with a limited set of attributes were routed to
SQE. Over time, more queries will use SQE, and increasingly fewer queries will use CQE. At
some point, all queries, or at least those that originate from SQL interfaces, will use SQE.
Note: SQE processes queries only from SQL interfaces, such as interactive and
embedded SQL, Open Database Connectivity (ODBC) and Java™ Database Connectivity
(JDBC™).
14
2.2.1 Database architecture before V5R2M0
For systems prior to the release of V5R2M0, all database requests are handled by the CQE.
Figure 2-3 shows a high-level overview of the architecture of DB2 Universal Database for
iSeries before OS/400 V5R2. The optimizer and database engine are implemented at
different layers of the operating system.
ODBC/JDBC/ADO/DRDA/XDA
Network
Host Server
Native
(Record I/O)
CLI/JDBC
Static
Dynamic
Extended
Dynamic
Compiled
embedded
statements
Prepare
everytime
Prepare once
and then
reference
SQL
Optimizer
Machine Interface (MI)
SLIC
DB2 Universal Database
(Data Storage and Management)
The optimizer and
database engine
are separated at
different layers of
the operating
system
Figure 2-3 Database architecture before the release of V5R2M0: Classic Query Engine
Most CQE query decisions are made above the machine interface (MI) level. In CQE, the
interaction between the optimizer and the query execution component occurs across the MI,
resulting in interface-related performance overhead.
2.2.2 Current database architecture
With the release of V5R2M0, a new SQE was shipped. SQE and CQE coexist in the same
database environment. Depending on the database requests, the Query Dispatcher (see
2.2.3, “Query Dispatcher” on page 17) decides to route the query to either the CQE or SQE.
While both the new SQE and the existing CQE can handle queries from start to finish, the
redesigned engine simplifies and speeds up queries. In addition to providing the same
functionality as CQE, SQE also performs these functions:
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
Moves the optimizer below the MI for more efficient query processing
Separates and moves improved statistics to the Statistics Manager dashboard
Uses an object-oriented design that accelerates the delivery of new database functionality
Uses more flexible, independent data access options to provide autonomous query cruise
control
Uses enhanced algorithms to provide greater responsiveness and query handling
Provides enhanced performance on long-running complex query terrains
Retains road maps to provide ease of use in query driving
Provides additional and enhanced query feedback and debug information messages
through the Database Monitor and Visual Explain interfaces
15
There are several new and updated components of SQE in OS/400 V5R2 and i5/OS V5R3,
including:
򐂰
򐂰
򐂰
򐂰
򐂰
Query Dispatcher
Statistics Manager
SQE Optimizer
Data Access Primitives
Plan Cache
Figure 2-4 shows an overview of the DB2 Universal Database for iSeries architecture on
i5/OS V5R3 and where each SQE component fits. The functional separation of each SQE
component is clearly evident. In line with design objectives, this division of responsibility
enables IBM to more easily deliver functional enhancements to the individual components of
SQE, as and when required.
ODBC/JDBC/ADO/DRDA/XDA
Network
Host Server
CLI/JDBC
Static
Dynamic
Extended
Dynamic
Compiled
embedded
statements
Prepare
every time
Prepare once
and then
reference
SQL
Query Optimizer
Native
(Record I/O)
Query Dispatcher
CQE Optimizer
SQE Optimizer
Machine Interface (MI)
DB2 Universal Database (Data Storage and Management)
SLIC
SQE Optimizer
CQE
Database Engine
The optimizer and
database engine
merged to form the
SQL Query Engine
and much of the
work was moved
to SLIC
SQL Statistics
Manager
SQE Primitives
Figure 2-4 Current database architecture: Coexisting CQE and SQE
Note: Most of the SQE Optimizer components are implemented below the MI level, which
translates into enhanced performance.
16
Object-oriented design
The SQE query optimizer was written using an object-oriented design and implementation
approach. It uses a tree-based model of the query, where each node is an independent and
reusable component. These components can interact and interface with each other in any
given order or combination. Each node can be optimized and executed independently. This
design allows greater flexibility when creating new methods for query implementation.
With this new design, the ability to use an index in reverse order can be easily implemented
by simply adding a new node. The procedural nature of CQE prevents it from being easily
enhanced to read an index in reverse order.
Another example to demonstrate how the object-oriented tree model makes SQE easier to
enhance is SQE support for nonsensical queries. The term nonsensical describes a query
statement that does not return any result rows, for example:
Select * from testtable where 1 = 0
Surprisingly, many applications use this type of query to force no rows to be returned.
Because of the procedural nature of CQE, it is virtually impossible to enhance CQE to
recognize the fact that 1 will never equal 0. Therefore, CQE implements this query using a
table scan. In contrast, the tree node model of SQE easily allows a node to be added to check
for nonsensical predicates before reading any rows in the specified table.
Figure 2-5 shows an example of the node-based implementation used in SQE. In this
example, NODE1 represents a typical index probe access method. New nodes to check for
nonsensical queries and to index in reverse order are added.
Original Design
Additional Nodes
Node 1
Index Probe
Node 4
Use index in
reverse order
Node 2
Bring index
into memory
Node 3
Test index
for predicate
Node 5
Check for nonsense
queries
Figure 2-5 Object-oriented design: Tree-based model of queries
2.2.3 Query Dispatcher
The function of the Query Dispatcher is to route the query request to either CQE or SQE,
depending on the attributes of the query. One of these attributes includes the interface from
which the query originates, which is either SQL-based (embedded SQL, ODBC, or JDBC) or
non-SQL based (OPNQRYF and Query/400). All queries, irrespective of the interface used,
are therefore processed by the dispatcher. It is not possible for a user or application program
to influence this behavior or to bypass the dispatcher.
Note: Only SQL queries are considered for the SQE. OPNQRYF and Query/400 are not
SQL based.
17
Figure 2-6 illustrates how different database requests are routed to the different query
engines.
Non-SQL Interfaces
QPNQRYF
Query/400
QQQQry API
SQL-Based Interfaces
ODBC/JDBC/CLI
Embedded and Interactive SQL
Run SQL Scripts
Query Manager
Net.Data
RUNSQLSTM
Optimizer
Query Dispatcher
CQE
SQE
Parts of the SQE Optimizer
now reside in SLIC
System Licensed Internal Code (SLIC)
CQE
DATA
SQE
Figure 2-6 Query Dispatcher routing database requests to the query engines
The staged implementation of SQE enabled a limited set of queries to be routed to SQE in
V5R2. In general, read-only single table queries with a limited set of attributes were routed to
SQE. With the V5R2 enabling PTF applied (PTF SI07650), the dispatcher routes many more
queries through SQE. More single table queries and a limited set of multi-table queries can
take advantage of the SQE enhancements. Queries with OR and IN predicates might be
routed to SQE with the enabling PTF as are SQL queries with the appropriate attributes on
systems with symmetric multiprocessing (SMP) enabled.
Note: For more information about PTF SI07650, see Informational APAR II13486 on the
Web at:
http://www-912.ibm.com/n_dir/nas4apar.nsf/042d09dd32beb25d86256c1b004c3f9a/61ffe88d56a94
3ed86256c9b0041fbeb?OpenDocument
In i5/OS V5R3, a much larger set of queries is implemented in SQE including those with the
enabling PTF on V5R2 and many queries with the following types of attributes:
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
Views
Sub-Selects
Common Table Expressions
Derived Tables
Unions
Updates
Deletes
STAR_JOIN (without FORCE_JOIN_ORDER)
SQL queries that continue to be routed to CQE in i5/OS V5R3 have the following attributes:
򐂰
򐂰
򐂰
򐂰
18
LIKE predicates
Large Objects (LOB)
Sensitive Cursor
NLSS/CCSID translation between columns
򐂰
򐂰
򐂰
򐂰
DB2 Multisystem
ALWCPYDTA(*NO)
References to logical files
Tables with select/omit logical files over them
Note: The QAQQINI option IGNORE_ DERIVED_INDEX allows SQE to process the query
even when a derived key or select/omit index exists over a table in the query. If allowed to
run, SQE ignores the derived and select/ omit indexes.
Derived keys occur when either the Create Logical File (CRTLF) command’s data
description specification (DDS) specifies keys that are derivations, for example Substring,
or if an NLSS sort sequence is active when the CREATE INDEX SQL statement is
performed. By default, if one of these indexes exists over a table in the query, SQE is not
allowed to process the query.
The dispatcher also has the built-in capability to reroute an SQL query to CQE that was
initially routed to SQE. A query typically reverts to CQE from SQE whenever the optimizer
processes table objects that define any of the following logical files or indexes:
򐂰 Logical files with the SELECT/OMIT DDS keyword specified
򐂰 Logical files built over multiple physical file members
򐂰 Nonstandard indexes or derived keys, such as logical files specifying the DDS keywords
RENAME or Alternate Collating Sequence (ACS) on a field referenced in the key
򐂰 Sort sequence NLSS specified for the index or logical file
Note: SQL requests that are passed back to CQE from SQE might experience an
overhead of up to 10 to 15% in the query optimization time. However, that overhead is not
generated every time that an SQL statement is run. After the access plan is built by CQE,
the Query Dispatcher routes the SQL request to CQE on subsequent executions. The
overhead appears when the access plan is built the first time or rebuilt by the optimizer.
2.2.4 Statistics Manager
In releases before V5R2, the retrieval of statistics was a function of the CQE Optimizer. When
the optimizer needed to know information about a table, it looked at the table description to
retrieve the row count and table size. If an index was available, the optimizer might then
extract further information about the data in the table. Figure 2-7 illustrates how CQE relies on
indexes for statistics.
19
Table Size
Row Count
Radix Index
Encoded
Vector Index
Statistics Retrieval
SQL
Query
CQE Optimizer
Access Plan
Above MI Level
Figure 2-7 CQE Optimizer and Statistics Retrieval
In V5R2, the collection of statistics was removed from the optimizer and is now handled by a
separate component called the Statistics Manager. The Statistics Manager does not actually
run or optimize the query. It controls the access to the metadata and other information that is
required to optimize the query. It uses this information to answer questions posed by the
query optimizer. The Statistics Manager always provides answers to the optimizer. In cases
where it cannot provide an answer based on actual existing statistics information, it is
designed to provide a predefined answer.
Note: This new statistical information is used only by the SQE. Queries that are dispatched
to the CQE do not benefit from available statistics, nor do they trigger the collection of
statistics.
Figure 2-8 shows the new design with Statistics Manager.
Table Size
Row Count
Radix Index
Encoded
Vector Index
Statistics
Manager
SQL
Query
SQE Optimizer
Access Plan
Below MI Level
Figure 2-8 SQE Optimizer and Statistics Manager
20
The Statistics Manager controls the access to the metadata that is required to optimize the
query. It uses this information to provide answers to the questions posed by the query
optimizer. The Statistics Manager typically gathers and keeps track of the following
information:
򐂰 Cardinality of values
This is the number of unique or distinct occurrences of a specific value in a single column
or multiple columns of a table.
򐂰 Selectivity
Also known as a histogram, this information is an indication of how many rows will be
selected by any given selection predicate or combination of predicates. Using sampling
techniques, it describes the selectivity and distribution of values in a given column of the
table.
򐂰 Frequent values
This is the top nn most frequent values of a column together with a count of how frequently
each value occurs. This information is obtained by using statistical sampling techniques.
Built-in algorithms eliminate the possibility of data skewing. For example, NULL values and
default values that can influence the statistical values are not taken into account.
򐂰 Metadata information
This includes the total number of rows in the table, which indexes exist over the table, and
which indexes are useful for implementing the particular query.
򐂰 Estimate of I/O operation
This is an estimate of the amount of I/O operations that are required to process the table
or the identified index.
You can obtain the majority of this information from existing binary-radix indexes or
encoded-vector indexes. An advantage of using indexes is that the information is available to
the Statistics Manager as soon as the index is created or maintained.
2.2.5 SQE Optimizer
Like the CQE Optimizer, the SQE Optimizer controls the strategies and algorithms that are
used to determine which data access methods should be employed to retrieve the required
data. Its purpose is to find the best method to implement a given query.
A fundamental characteristic distinguishes the SQE Optimizer from the CQE Optimizer. The
SQE Optimizer gets access to the statistic data collected by the Statistic Manager, by simply
asking questions related to the system and the tables used in the query. Based on this
information, the access method is determined based on the one that has the least amount of
CPU utilization and I/O overhead costs.
Because most of the SQE Optimizer functionality is implemented beneath the MI and
consequently closer to the data, the database management system allows for greater
flexibility and increased performance.
The CQE Optimizer uses a clock-based timeout algorithm. The CQE Optimizer resequences
the indexes, based on the number of matching index columns and the operators used in the
WHERE clause of the SQL statement. This approach ensures that the most efficient indexes
are optimized first, before the set time limit expired.
In contrary, the amount of time that the SQE Optimizer spends optimizing an access plan is
unlimited. A check is done to determine if any indexes exist on the table, with keys built over
the columns specified in WHERE or SELECT clauses of the SQL statement. These indexes
21
are then resequenced so that the most appropriate indexes are processed first and
reorganized further based on index-only access, index probe selectivity, total index selectivity,
and the size of the index keys.
Note: For SQE, the indexes are ordered in general so that the indexes that access the
smallest number of entries are examined first. For CQE, the indexes are generally ordered
from mostly recently created to oldest.
2.2.6 Data Access Primitives
The basic function of SQE Data Access Primitives is to implement the query. Using the data
access methods derived from the object-oriented, tree-based architecture, Data Access
Primitives provide the implementation plan of the query.
2.2.7 Access plan
The methods used for a specific SQL statement to get access to the data are stored in access
plans. If the access plan does not exist, it is created the first time that an SQL statement is
executed. If the access plan already exists, it is compared with the information provided by
the Statistics Manager (SQE) or by the query optimizer in CQE. If the optimizer decides to
use an other access path, the access plan is updated.
In contrary to CQE, the access plans that are created and used with the SQE are organized in
a tree-based structure to allow for maximum flexibility.
If you use SQL statements in programs, there are different ways to embed, prepare, and
execute your SQL statements. These different methods affect the creation time of the access
plan for the specified SQL statements. All executable SQL statements must be prepared
before they can be executed. The result of preparation is the executable or operational form of
the statement. We can differentiate between the three methods:
򐂰 Static SQL
In static SQL, the SQL statements that must be executed are already known at compile
time. The precompiler checks the syntax and converts the SQL statement into an
executable form, as well as creates an access plan that is embedded into the program
object. If the access plan is changed because of an altered data pool or new indexes, the
access plan is updated in the program object. In this way, a program object can grow over
the time, even if no modifications are performed.
򐂰 Dynamic SQL
Programs that contain embedded dynamic SQL statements must be precompiled like
those that contain static SQL. Unlike static SQL, the dynamic SQL statements are
checked, constructed, and prepared at run time. The source form of the statement is a
character or graphic string that is passed to the database manager by the program that is
using the static SQL PREPARE or EXECUTE IMMEDIATE statement. The operational
form of the statement persists for the duration of the connection or until the last SQL
program leaves the call stack. Access plans associated with dynamic SQL might not
persist after a database connection or job is ended.
򐂰 Extended dynamic SQL
An extended dynamic SQL statement is neither fully static nor fully dynamic. The Process
Extended Dynamic SQL (QSQPRCED) API provides users with extended dynamic SQL
capability. Like dynamic SQL, statements can be prepared, described, and executed using
this API. Unlike dynamic SQL, SQL statements prepared into a package by this API
persist until the package or statement is explicitly dropped.
22
The iSeries Access ODBC driver and JDBC driver both have extended dynamic SQL
options available. They interface with the QSQPRCED API on behalf of the application
program.
SQL packages
SQL packages are permanent objects with the object type *SQLPKG used to store
information related to prepared, extended dynamic SQL statements. They can be used by the
iSeries Access for Windows® ODBC driver and the IBM Toolbox for Java JDBC driver. They
are also used by applications which use the QSQPRCED API interface.
The SQL package contains all the necessary information to execute the prepared statement.
This includes registry of the statement name, the statement text, the internal parse tree for
the statement, definitions of all the tables and fields involved in the statement, and the query
access plan needed to access the tables at run time.
Note: When using embedded SQL, no separate SQL package is created, but the access
plan is integrated into the program or service program object.
Creation time of SQL packages
In the case of ODBC and JDBC, the existence of the package is checked when the client
application issues the first prepare of an SQL statement. If the package does not exist, it is
created at that time, even though it might not yet contain any SQL statements. In the case of
QSQPRCED, creation of the package occurs when the application calls QSQPRCED
specifying function 1.
Advantages of SQL packages
Because SQL packages are a shared resource, the information built when a statement is
prepared is available to all the users of the package. This saves processing time, especially in
an environment when many users are using the same or similar statements. Because SQL
packages are permanent, this information is also saved across job initiation or termination
and across initial program loads (IPLs). In fact, SQL packages can be saved and restored on
other systems. By comparison, dynamic SQL requires that each user go through the prepare
processing for a particular statement and do this every time the user starts the application.
SQL packages also allow the system to accumulate statistical information about the SQL
statements. Accumulating such information results in better decisions about how long to keep
cursors open internally and how to best process the data needed for the query. As indicated
previously, this information is shared across users and retained for future use. In the case of
dynamic SQL, every job and every user must relearn this information.
Deletion of SQL packages
Packages must be deleted when the underlying metadata for statements stored in the
package has been changed. If a table, view, procedure, or other SQL object is altered, the
information in the package is not updated. If the package is not deleted, a variety of unusual
errors can occur, including truncation, data mapping errors, incorrect describe information,
and so on.
Delete packages whenever significant changes (those that might cause a large amount of
access plan rebuilds) are made to the database, operating system, or hardware. Because
extended dynamic SQL packages are recreated when the application is run, there is little
harm in deleting them.
23
Plan Cache
The Plan Cache is a repository that contains query implementation plans for queries
optimized by the SQE Optimizer. Query access plans generated by CQE are not stored in the
Plan Cache. The architecture of DB2 Universal Database for iSeries allows for only one Plan
Cache per iSeries server or logical partition (LPAR).
The purpose of the Plan Cache is to facilitate the reuse of a query access plan at some future
stage when the same query, or a similar query, is executed. After an access plan is created, it
is available for use by all users and all queries, irrespective of the interface from which the
query originates.
CQE already uses plan caches, but queries from different interfaces each go to their own
Plan Cache. Furthermore, when an access plan is tuned, for example when creating an
index, all queries can benefit from this updated access plan. The update plan eliminates the
need to reoptimize the query and results in greater efficiency and faster processing time. In
the case of CQE, each query has to update its own access plan to benefit from the newly
created index.
Before optimizing an incoming query, the optimizer looks for the query in the plan cache. If an
equivalent query is found, and the associated query plan is found to be compatible with the
current environment, the already-optimized plan is used, avoiding full optimization.
Figure 2-9 shows the concept of reusability of the query access plans stored in the Plan
Cache. The Plan Cache is interrogated each time a query is executed using the SQE.
SQL Pgm-A
Plan Cache
Statement 1
Plan X
Statement 2
SQL PKG-1
Statement 3
Statement 4
Plan Y
CQE
Access plans are not
stored in the Plan Cache
SQL PKG-2
Plan Z
Statement 3
Statement 5
Figure 2-9 Plan Cache
Note: Access plans generated by CQE are not stored in the SQE Plan Cache.
In addition, unlike CQE, SQE can save multiple different plans for the same query. This
method is useful in more dynamic environments where the plan changes depending on user
inputs, available memory, and so on.
Note: The Plan Cache is cleared during an IPL or varying the independent auxiliary
storage pool (IASP).
24
For more information about SQE and CQE, see the IBM Redbook Preparing for and Tuning
the V5R2 SQL Query Engine on DB2 Universal Database for iSeries, SG24-6598, and DB2
Universal Database for iSeries Database Performance and Query Optimization, which is
available in the iSeries Information Center on the Web at:
http://publib.boulder.ibm.com/infocenter/iseries/v5r3/index.jsp?topic=/rzajq/rzajqmst.htm
2.3 Star Join Schema
A star schema is a database model characterized by a large centralized table, called the fact
table, surrounded by other highly normalized tables called dimension tables. A Star Join
Schema query is a query over multiple tables of a star schema database with local selection
specified on the dimension or dimensions and equi-join predicates between the fact table and
the relevant dimension table or tables.
The attributes of a star schema database model usually include:
򐂰 A relatively large fact table that contains millions or billions of rows holding the measurable
or additive “facts” such as sales type transactions or events
򐂰 Relatively small and highly normalized dimension tables that contain descriptive data
about the “facts” (in the central fact table), such as customer or location information
򐂰 A central fact table that depends on the surrounding dimension tables using a parent/child
relationship, with the fact table as the child and the dimension tables as the parent
If the dimension tables are further normalized, the results are dimensions that might have
additional tables that support them. This is known as a “snowflake” schema or model.
Figure 2-10 shows the structure of a typical Star Join Schema.
Dimension Table
Customer
Dimension Table
Dimension Table
Product
Fact Table
Supplier
Sales
Dimension Table
Dimension Table
Geography
Time
Figure 2-10 Star Join Schema
Starting with i5/OS V5R3, the SQE automatically supports optimizing Star Join Schema
queries. This new support does not require the use of QAQQINI options. Recognizing and
optimizing a star schema query is now a normal part of handling any query request.
25
Support is provided with DB2 Universal Database for iSeries CQE in OS/400 V5R2 and with
CQE and SQE in i5/OS V5R3. To enable Star Join Schema support via QAQQINI options file,
perform one of the following operations prior to running the query:
INSERT INTO library/QAQQINI VALUES('STAR_JOIN', '*COST', NULL);
INSERT INTO library/QAQQINI VALUES('STAR_JOIN', '*FORCE', NULL);
In the first operation, *COST is the recommended and preferred parameter value, since each
EVI is considered based on its potential to further minimize the I/O. In the second operation,
*FORCE is more aggressive, using a larger number of bitmaps with the CQE skip sequential
access method, since all of the relevant EVIs are used for local selection, regardless of cost.
With SQE, the QAQQINI option STAR_JOIN is ignored. However, because a few queries
might still be routed to CQE, users who have specified STAR_JOIN *COST in the past should
continue to do so. That way CQE will not suffer and SQE will be unaffected. In addition, in the
past, the FORCE_JOIN_ORDER option was specified in conjunction with the STAR_JOIN
option. This option is no longer required for CQE and should be removed by customers who
use it for star join queries. If this option is left in, the FORCE_ JOIN_ORDER is still honored
by both CQE and SQE and might prevent the optimizer from generating optimal plans.
Note: Single key EVIs created over the foreign key columns of the fact table are required
for the CQE Optimizer to implement the star join techniques. These same EVIs are optimal
for the new SQE Optimizer to implement its Star Join Schema techniques.
2.3.1 Queries in a Star Join Schema
Queries in a Star Join Schema typically include the following characteristics:
򐂰 Multiple tables participating in the query
򐂰 Local selection predicates on the dimension tables rather than the larger fact table
򐂰 Equi-join predicates between the dimension tables and the fact table used to locate and
select the relevant fact table rows and to decode and describe the fact table data
The equi-join predicate between any one dimension table and the fact table might result in
a very large number of fact table rows being selecting, while the intersection of the
equi-join predicates of multiple dimension tables might result in a relatively small number
of fact table rows being selected.
To achieve the optimal performance in joining the tables, you must consider these rules:
򐂰 For CQE, the join order of the query must be Fact_Table as the primary table, followed by
the dimension tables. CQE identifies the fact table by determining the largest table in the
query and that table is “pinned” in join position 1. Then the order of dimension tables
(2 through n) is optimized.
򐂰 SQE also identifies the largest table in the query (that is, the fact table) but optimizes the
join order of all the tables based on new strategies and methods. This might result in the
fact table being placed somewhere other than the first join position. Since SQE can use
new and different methods, it is advantageous to also create single key radix indexes on
the foreign key columns of the fact table. This index is optimal if the fact table is being
joined from a dimension table in join position 1.
26
2.3.2 Restrictions and considerations
When working with Star Join Schema support, consider these restrictions:
򐂰 For both CQE and SQE, the only sort sequence supported is *HEX. If the query or job
specifies a sort sequence other than *HEX, the Star Join Schema support is not used.
򐂰 Single key column EVIs must be created over the join columns of the fact table. When
using the star schema support, the CQE Optimizer does not use radix indexes to create
the dynamic bitmaps for the skip sequential processing on the fact table. When a large
range of key values are selected, using EVIs to create the dynamic bitmaps is more
efficient and faster.
򐂰 Specifying the QAQQINI parameter STAR_JOIN within an OS/400 V5R2 database
environment causes SQL query requests to use CQE instead of SQE. In V5R2, SQE is
not specifically enhanced for Star Join Schema queries.
In V5R3, SQE includes specific enhancements for optimizing and executing Star Join
Schema queries, so the STAR_JOIN parameter is ignored by SQE. In V5R3, CQE
continues to honor the STAR_JOIN parameter for those query requests that are optimized
and executed by CQE. For example, a query that contains the LIKE operator does
continue to be optimized and executed by CQE instead of SQE. To gain the benefit of the
CQE star join support, the QAQQINI STAR_JOIN parameter is needed.
2.3.3 Lookahead Predicate Generation
The key feature of the new SQE support is referred to as Lookahead Predicate Generation
(LPG). While the SQE LPG support looks similar to the older CQE support, it has some
additional benefits and uses.
When a query in a Star Join Schema is optimized, a hash join is normally chosen as the
method to join the fact table to the dimension tables. To accomplish this, the original query is
broken into multiple parts or steps. For each dimension table, an internal query is executed to
access the rows matching the local selection predicates, using the best available access
method (scan or probe, with or without parallelism). The data required for the query is then
used to build a hash table.
In addition to building the hash tables, the join key values of the selected dimension table
rows are used to populate a list of distinct keys. This list represents all the join key values (that
match the corresponding local selection) and is used to identify the join column values in the
fact table.
After the hash table or tables and distinct key list or lists are built, the original query is
rewritten. The distinct key list or lists are used to provide local selection on the fact table. In
effect, it transfers the local selection from the dimension table or tables to the fact table. This
transfer is referred to as Lookahead Predicate Generation.
For more information about the Star Join Schema and LPG, see the white paper Star Schema
Join Support within DB2 UDB for iSeries - Version 3 on the Web at:
http://www-1.ibm.com/servers/enable/site/education/abstracts/16fa_abs.html
For more information about the SQE, see the IBM Redbook Preparing for and Tuning the
V5R2 SQL Query Engine on DB2 Universal Database for iSeries, SG24-6598.
27
28
Part 2
Part
2
Gathering, analyzing,
and querying database
performance data
In this part, we describe and discuss the different ways to gather database performance data.
After we explain how to gather this data, we describe the different ways to analyze it with the
tools that DB2 Universal Database for iSeries has. Later in the part, we show you how to
query the Database performance data and how to tie this data to one of the preferred tools
called Visual Explain.
This part contains the following chapters:
򐂰 Chapter 3, “Overview of tools to analyze database performance” on page 31
򐂰 Chapter 4, “Gathering database SQL performance data” on page 51
򐂰 Chapter 5, “Analyzing database performance data using iSeries Navigator” on page 93
򐂰 Chapter 6, “Querying the performance data of the Database Monitor” on page 133
򐂰 Chapter 7, “Using Collection Services data to identify jobs using system resources” on
page 179
򐂰 Chapter 8, “Analyzing database performance data with Visual Explain” on page 197
29
30
3
Chapter 3.
Overview of tools to analyze
database performance
Database performance is a high priority in any system. The objective is to maximize system
resource utilization, while achieving maximum performance throughput. Therefore, analyzing
your queries is the most important step to ensure that they are tuned for optimal performance.
You must select the proper tools for collecting and analyzing the database performance data
first, to ensure that your queries are tuned for optimal performance. By using the following
analysis tools to obtain an information through the monitoring processes, you should be able
to take the appropriate corrective actions.
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
Current SQL for a Job (iSeries Navigator)
Print SQL information
Debug messages
Index Advisor
Index Evaluator
SQL Performance Monitor (Detailed Monitor)
Memory-based Database Monitor (Summary Monitor)
Visual Explain
In this chapter, we introduce and provide information about the tools for monitoring and
analyzing the database performance data of your SQL queries.
31
3.1 Current SQL for a Job function in iSeries Navigator
You can use the Current SQL for a Job function to select any job running on the system and
display the current SQL statement being run, if any. In addition to displaying the last SQL
statement being run, you can edit and rerun it through the Run SQL Script option (linked
automatically) and display the actual job log for the selected job or, even end the job. You can
also use this function for database usage and performance analysis with the Visual Explain
tool as explained in Chapter 8, “Analyzing database performance data with Visual Explain” on
page 197.
To start the Current SQL for a Job function, in iSeries Navigator, in the left pane, right-click
Databases and select Current SQL for a Job as shown in Figure 3-1.
Figure 3-1 Selecting the Current SQL for a Job function in iSeries Navigator
Then the Current SQL window (Figure 3-2) opens. This window displays the name, user, job
number, job subsystem, and current user for the available jobs on your system. You can
select a job and display its job log, the SQL statement currently being run (if any), decide to
reuse this statement in the Run SQL Script Center, or end the job, provided that you have
sufficient authority.
Note: The line “Last statement to finish as of 4:33:18 PM” is the time that the SQL
statement button was selected in iSeries Navigator. This is not an indication of when the
SQL ran. The SQL statement displayed might be the current SQL the job is running or the
last SQL statement ran in the job (7).
32
2
4
1
7
5
6
3
Figure 3-2 Current SQL for a job
In our example, we selected a Java Database Connectivity (JDBC) job (1) and clicked the
SQL Statement button (2) to view the last SQL statement that it ran, in the bottom part of the
panel (3). To go to its job log, we can click the Job Log button (4). After the SQL statement is
displayed in the bottom part of the panel, we can click the Edit SQL button (5) to work on this
same statement with the Run SQL Script center as shown in Figure 3-3.
Figure 3-3 Run SQL Script with Current SQL for a job
Chapter 3. Overview of tools to analyze database performance
33
We can click the Run Visual Explain button (6) to perform direct analysis of the SQL
statement as shown in Figure 3-4.
Refer to Chapter 8, “Analyzing database performance data with Visual Explain” on page 197,
for a more detailed explanation of how to analyze a query and use the functionality in Visual
Explain within iSeries Navigator.
Figure 3-4 Running Visual Explain in Current SQL for a job
3.2 Print SQL information
The information contained in SQL packages, service programs, and embedded SQL
statements can also assist in identifying potential performance problems in your queries.
To view the information pertaining to the implementation and execution of your query, select
an SQL package from iSeries Navigator. Right-click the SQL package name and select
Explain SQL, as shown in Figure 3-5.
This is equivalent to using the Print SQL Information (PRTSQLINF) CL command that
extracts the optimizer access method information from the object and places that information
in a spool file. The spool file contents can then be analyzed to determine if any changes are
needed to improve performance.
The information in the SQL package is comparable to the debug messages discussed in 3.3,
“Debug messages” on page 36. However, there is more detail in the first level SQLxxxx
messages.
34
Figure 3-5 Selecting to print SQL information
Figure 3-6 shows an example of the SQL package information that can be displayed. As you
can see in the example, the information pertaining to the join order of the tables, the access
methods used in the implementation of the query, and runtime statistics are available. Notice
the messages that indicate which QAQQINI query options file was used for executing the
query and whether this query implementation used symmetric multiprocessing (SMP).
Figure 3-6 Viewing SQL package information
You can also obtain this information by using the following PRTSQLINF CL command:
PRTSQLINF OBJ(library_name/program_name or package_name) OBJTYPE(*SQLPKG)
Note: If you are unsure of the program or package name, look at the program name above
the QSQ* module in the call stack when you use option 11 (Display call stack, if active) of
the Work with Active Jobs (WRKACTJOB).
The PRTSQLINF CL command directs the output data to a spooled file, from where you can
display or print the information.
35
Note: The information retrieved from an SQL package might not accurately reflect the
access plan used by the last execution of the SQL query. For example, if circumstances at
execution time cause the query to be reoptimized and the package or program is locked,
then a new access plan is dynamically generated and placed in the Plan Cache (for SQL
Query Engine (SQE) use). The version stored in the package or program is not updated.
3.3 Debug messages
Analyzing debug messages is another important tool for monitoring and tuning queries. When
doing so, keep in mind the following points:
򐂰 The debug messages are no longer being enhanced (that is, no new messages are being
added for queries that go through SQE).
򐂰 It is hard to tie a message to an SQL statement.
򐂰 It is difficult to search through all of the job log messages.
There are multiple methods of directing the system to generate debug messages while
executing your SQL statements such as:
򐂰
򐂰
򐂰
򐂰
Selecting the option in the Run SQL Scripts interface of iSeries Navigator
Using the Start Debug (STRDBG) CL command
Setting the QAQQINI table parameter
Using Visual Explain
You can choose to write only the debug messages for one particular job to its job log. If you
want to use iSeries Navigator to generate debug messages, in the Run SQL Scripts window,
click Options → Include Debug Messages in Job Log as shown in Figure 3-7.
Figure 3-7 Enabling debug messages for a single job
36
After you run your query, in the Run SQL Scripts window, select View → Joblog to view the
debug messages in the job log. In our example, we used the SQL statement shown in
Example 3-1.
Example 3-1 Example SQL statement
SELECT year, month, returnflag, partkey, quantity, revenue_wo_tax
FROM veteam06.item_fact
WHERE year = 2002 and month = 6 and returnflag = 'R';
The detailed job log describes information that you can use to identify and analyze potential
problem areas in your query such as:
򐂰
򐂰
򐂰
򐂰
򐂰
Indexes
File join order
Temporary result
Access plans
Open data paths (ODPs)
All of this information is written to the job log when under debug using the STRDBG
command.
Figure 3-8 shows an example of the debug messages contained in the job log after you run
the previous query.
Figure 3-8 Job log debug messages
After you enable these settings for a particular job, only debug messages relating to queries
running in that job are written to the job log. You see the same debug messages with this
option as those explained later when the QAQQINI parameter MESSAGES_DEBUG is set to
*YES. You also see additional SQL messages, such as “SQL7913 - ODP deleted and SQL7959
- Cursor CRSRxxxx closed”, in the job log.
Select any of the debug messages displayed in the job log. Click File → Details to obtain
more detailed information about the debug message. Figure 3-9 shows an example of a
detailed debug message that is displayed.
By looking at the messages in the job log and reviewing the second-level text behind the
messages, you can identify changes that might improve the performance of the query such
as:
򐂰
򐂰
򐂰
򐂰
Why index was or was not used
Why a temporary result was required
Join order of the file
Index advised by the optimizer
37
Figure 3-9 Detailed debug message information
When running SQL interactively, either through a 5250-session or via the Run SQL Scripts
window in iSeries Navigator, you can also use the STRDBG CL command to generate debug
messages in your job log.
Remember: You also must run the Start Server Job (STRSRVJOB) CL command if your
query runs as a batch job.
By setting the value of the QAQQINI parameter MESSAGES_DEBUG to *YES, you can direct
the system to write detailed information about the execution of your queries into the job’s job log.
To activate this setting through the Run SQL Scripts interface of iSeries Navigator, select the
QAQQINI table that you want to use as shown in Figure 3-10. Then, select the
MESSAGES_DEBUG parameter and change the parameter value as shown in Figure 3-10.
After you make the appropriate change, close the window, and in the Change Query
Attributes window, click OK to save your changes.
Figure 3-10 Enabling debug messages in QAQQINI
38
Important: Changes made to the QAQQINI table are effective immediately. They affect all
users and queries that use this table. For example, if you set the MESSAGES_DEBUG
parameter to *YES in a particular QAQQINI table, all queries that use that QAQQINI table
write debug messages to their respective job logs.
The analysis of optimizer debug messages was made easier with the addition of a Predictive
Query Governor. By specifying a time limit of zero in the Predictive Query Governor, query
optimizer debug messages can be generated in the job log without running the query. The
query time limit is checked against estimated query time before initiating your query since the
optimization cost and access plan are determined prior to execution in cost-based
optimization.
The time limit is set on a per-job basis such as:
򐂰 The QRYTIMLMT parameter on the CHGQRYA CL command
򐂰 The QUERY_TIME_LIMIT parameter in the QAQQINI file
򐂰 The QQRYTIMLMT system value (CHGSYSVAL QQRYTIMLMT)
That is, you can analyze a query, which might take 16 hours to run, in only a few seconds.
Some changes can be made to the query or to the database. The effect can be modeled on
the query in a few minutes. The query is then run when the optimum implementation is
achieved.
One of the most important debug messages to look for is advise about the creation of
indexes, since the objective of creating indexes is to improve the performance of your queries.
The query optimizer analyzes the record selection in the query and determines, based on the
default estimate, whether the creation of an index can improve performance. If a permanent
index is beneficial, it returns the key fields necessary to create the suggested index. You can
find Index Advisor information in the debug message CPI432F. This takes us to the next tool,
which is the Index Advisor.
3.4 Index Advisor
In V5R3, the Index Advisor assists you more in suggesting indexes because the index
advised code was improved to recommend more useful indexes. To take advantage of this
enhancement, you are required to apply program temporary fix (PTF) MF34412.
Note: The improved Index Advisor is only for SQL requests that are routed to the SQE.
Index Advisor offers the following improvements:
򐂰 Advanced Radix Index advice
– It provides advice for a more optimal index, even when a suboptimal index exists and is
potentially used. For example, the query has two predicates, but an index exists on
only one of the predicates.
Note: This advanced function can increase the possibility of recommending an
unnecessary index where a suboptimal index is sufficient. Therefore, you must
analyze the recommendations that the advisor makes.
39
– It improves the handling of advice regarding join predicate, grouping, ordering, and
distinct clauses. A more complex combination of the record selection criteria can be
advised.
– Advise is based on a high level view of the query rather than the implementation
chosen.
򐂰 Advice for an encoded-vector index (EVI)
– EVI recommendations are made for certain grouping queries.
– Recommendations for the EVI are made when Lookahead Predicate Generation (LPG)
predicates are generated. For more information about LPG predicates, refer to 2.3.3,
“Lookahead Predicate Generation” on page 27.
In the past, you might have received the CPI432F message, which advised the use of an
index when an index already existed in key columns. Now the SQE Optimizer does not advise
an index if one already exists, even when the existing index differs from the advised index.
This occurs as a result of the one of the following reasons:
򐂰 The keys of the existing index match the keys of the advised index.
򐂰 The keys of the existing index are in a different, but acceptable, order than the advised
index.
Note: There are times when the keys have to be in the same order, such as in sorting,
so the SQE Optimizer advises an index if the existing one doesn’t match the correct
order.
򐂰 The order of the keys in the existing index are opposite of the advised index. For example,
the existing index is in descending order, and the advised index wants the keys in
ascending order.
3.5 Index Evaluator
Prior to V5R3, there was no easy way to determine if there were any unnecessary indexes
over a physical file or table. No information specified whether an index was used by a query or
when an index was used to give statistics to the query optimizer. The only information
available was from the Last Used Date. Unfortunately, customers erroneously used the date
to determine whether an index was used recently, and if it wasn’t used, they deleted the index.
After the indexes were deleted, performance problems resulted because the optimizer no
longer had the necessary indexes over the table for the query optimizer to use to make a
good decision. In V5R3 iSeries Navigator and V5R3 i5/OS, the Index Evaluator was added to
give statistics about index usage.
Note: To activate this feature, apply the PTFs for APAR SE14709, iSeries Access Service
Pack 2 for V5R3, and PTF SI14782.
The new statistics fields support the following characteristics:
򐂰 There are two counters, one for when the index is used in the query implementation and
one for when the index was used to gather statistics.
򐂰 Both counters are set by the two query engines, SQE and Classic Query Engine (CQE).
The statistic fields are updated regardless of the query interface, such as Query/400,
SQL/400®, OPNQRYF, or QQQQRY API, that is used.
򐂰 The statistics survive IPLs.
40
򐂰 Save/Restore does not reset the statistics on an index if an index is restored over an
existing index.
Note: If an index is restored that does not exist on the system, the statistics are reset.
򐂰 The statistics start counting after the PTFs are applied and active.
򐂰 For each counter, there is a corresponding time stamp to go with it from the last time the
counter was bumped.
򐂰 The data is stored internally in the index object. At this time, there is no way to query this;
to determine usage of an index over a specific time frame, look at each time stamp on
each individual index.
The new statistics are query statistics only. Native RPG, COBOL, and similar OPEN
operations that are not queries are not covered by the new statistic fields. However, we have
had native OPEN usage statistics for many years; if a user wants to determine whether the
index is used via these nonquery interfaces, look at the existing statistics via fields such as
Object Last Used.
Note: You can also use the QUSRMBRD API to return statistics.
To find this information, in iSeries Navigator, right-click the desired table and then click
INDEXES. The additional information is displayed as shown in Figure 3-11.
Figure 3-11 Getting index information in iSeries Navigator
41
In this example, you can see two indexes in the ITEM_FACT table as shown in Figure 3-12.
Figure 3-12 Two indexes in the ITEM_FACT table
The two counters update the columns:
򐂰
򐂰
򐂰
򐂰
Last Query Use
Last Query Statistic Use
Query Use Count
Query Statistics Use
Last Query Use is the column with the time stamp of when the index was used to access
tables in a query, and Query Use Count is updated with a corresponding count. Last Query
Statistics Use is updated when the optimizer uses the index to gather statistical information
on a table, and Query Statistics Use is updated with a corresponding count as shown in
Figure 3-13.
Figure 3-13 New information about indexes in V5R3 iSeries Navigator
42
To see the other information, such as an index definition and description, right-click the
indexes. Figure 3-14 shows an example of a detailed index description.
Figure 3-14 The detailed index descriptions
Note: You can also view the index information by using the Display File Description
(DSPFD) command on the table. But the information does not include the newly added
fields in V5R3.
3.6 The Database Performance Monitors
The Database Performance Monitors are integrated tools used to collect database-specific
performance information for SQL requests being run on the iSeries. They let you keep track of
resources that SQL statements use. The collected data is output to a database table or
tables. Then reports or queries are run against the collected data from these output tables to
analyze query optimization and the performance behavior of the database request. This
analysis helps to identify and tune performance problem areas.
Note: The Database Performance Monitor name is also known as SQL Performance
Monitor. From this point forward in the book, we use “SQL Performance Monitor” to refer to
the iSeries Navigator function. We use “Database Monitors” when using a green screen
command.
43
The SQL Performance Monitors can collect information about non-SQL query interfaces,
such as OPNQRYF and Query for iSeries, but their primary usage is for SQL-based
interfaces such as embedded SQL, Open Database Connectivity (ODBC), and JDBC.
The SQL Performance Monitors provide all the information that the STRDBG or PRTSQLINF
commands provide plus additional information such as:
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
System and job name
SQL statement text
Start and End time stamp
Estimated processing time
Total rows in table queried
Number of rows selected
Estimated number of rows selected
Estimated number of joined rows
Key columns for advised index
Total optimization time
ODP implementation
QAQQINI settings
The collected information and statistics can be analyzed later to determine such information
as:
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
The number of queries that rebuild access plans
The number of temporary indexes that have been created over a particular table
The queries that are the most time consuming
The user that is has the longest running queries
The queries that were implemented using reusable ODPs
Whether the implementation of a particular query changed with the application of a PTF or
a new release (uses a before and after comparison of monitor data)
Two types of SQL Performance Monitors are available with OS/400 and i5/OS, both which are
packaged with the operating system:
򐂰 Detailed Monitor
򐂰 Summary Monitor, also known as the Memory-based Database Performance Monitor
We discuss both monitors in the following sections.
3.6.1 Detailed Monitor
The Detailed Monitor gathers information about SQL requests and non-SQL queries, such as
OPNQRYF. It has details about optimization and runtime behavior. The monitor data is
dumped into a single output table.
Important: Use care when running the Detailed Monitor for long periods of time. The
collection and output of performance data can consume both disk and CPU resources with
its collection, resulting in system overhead.
STRDBMON and ENDDBMON
The CL commands Start Database Monitor (STRDBMON) and End Database Monitor
(ENDDBMON) provide the interface for the Detailed Performance Monitors. The commands
start and end the collection of database performance statistics.
The STRDBMON command gathers information about a query in real time and stores this
information in an output table. The information can help you determine whether your system
and your queries are performing as they should or whether they need fine tuning. The monitor
44
can be specified for a specific job or all jobs on system. Collected information and statistics
are placed in a single output database table made up of different record types.
Consider following parameters before you start database monitoring:
򐂰 OUTFILE parameter
– The file name is required; the library name is optional.
– It is created if it doesn’t exist or is reused if it does exit.
򐂰 OUTMBR parameter
– Defaults to first member in file
򐂰 JOB parameter
– Defaults to job issuing the STRDBMON command
– Can specify one job or *ALL jobs (no subsetting allowed)
򐂰 TYPE parameter
– The type of data collected is *SUMMARY(default) or *DETAIL.
– *SUMMARY provides all necessary analysis data.
– *DETAIL collects the same data as *SUMMARY plus the 3019 row.
• It causes a little more overhead.
• The Detailed SQL Performance Monitor interface in iSeries Navigator uses value of
type (*DETAIL).
Note: The *SUMMARY option in the TYPE parameter has no relationship to the
detailed and summary SQL Performance Monitors in iSeries Navigator.
򐂰 FRCRCD parameter
– This parameter uses the default value of *CALC.
– A larger number reduces the overhead of the monitor; a smaller number increases it.
򐂰 COMMENT parameter: Description of collection
The ENDDBMON command ends the Database Monitor for a specific job or all jobs on the
server. If an attempt to end the monitor on all jobs is issued, there must have been a previous
STRDBMON issued for all jobs. If a particular job is specified on this command, the job must
have the monitor started explicitly and specifically on that job.
When collecting information for all jobs, the Database Monitor collects on previously started
jobs or new jobs that are started after the monitor starts. Each job in the system can be
monitored concurrently by two monitors:
򐂰 One monitor started specifically on that job, and another started for all jobs in the system.
򐂰 When monitored by two monitors and each monitor is logging to a different output file,
monitor information is written to both files.
Database Monitors can generate significant CPU and disk storage overhead when in use. You
can gather performance information for a specific query instead of every query to reduce this
overhead on the system. You can also gather only specific monitoring data for smaller monitor
collection since large monitor database files can slow analysis.
Consider the following guidelines:
򐂰 If possible, try collecting data only for the job that you want.
򐂰 Collect monitor data only for long running SQL statement based on the optimizer’s
estimated runtime.
򐂰 Eliminate the SQL statement generated by DB2.
45
򐂰 By using the table filter function, collect only statements that reference certain tables such
as the one shown in the following example:
STRDBMON OUTFILE(o) COMMENT(‘TABLEFILTER(lib1/tab1,lib2/tab2)’)
Note: To use the TABLE FILTER function, apply the following required PTFs in advance.
򐂰 V5R2 PTFs: SI15035, SI15139, SI15140, SI15142, SI15143, SI15154, and SI15155
򐂰 V5R3 PTFs: SI15955, SI16331, SI16332, and SI16333
For more information about valid parameter settings for the QAQQINI table, refer to
“Monitoring your queries using Start Database Monitor (STRDBMON)” in DB2 Universal
Database for iSeries Database Performance and Query Optimization, which is available in the
iSeries Information Center on the Web at:
3.6.2 Summary Monitor or Memory Resident Database Monitor
The Memory Resident Database Monitor is a tool that provides another method for monitoring
database performance. This tool is intended only for the collection of performance statistics in
SQL queries. To collect performance statistics for non-SQL queries, you should start a
Detailed Database Monitor as explained in previous sections. The Memory-based Database
Monitor or Summary Monitor, with the help of a set of APIs, manages database monitoring
information for the user in memory. This memory-based monitor reduces CPU overhead and
resulting table sizes.
The STRDBMON command can constrain server resources when collecting performance
information. This overhead is mainly attributed to the fact that performance information is
written directly to a database table as the information is collected. The memory-based
collection mode reduces the server resources consumed by collecting and managing
performance results in memory. This allows the monitor to gather database performance
statistics with a minimal impact to the performance of the server as whole (or to the
performance of individual SQL statements).
The Summary Monitor collects much of the same information as the Detailed Database
Monitor, but the performance statistics are kept in memory. At the expense of some detail,
information is summarized for identical SQL statements to reduce the amount of information
collected. The objective is to get the statistics to memory as fast as possible while deferring
any manipulation or conversion of the data until the performance data is dumped to a result
table for analysis.
The Summary Monitor is not meant to replace the Detailed Database Monitor. There are
circumstances where the loss of detail in the SQL Performance Monitor is not sufficient to
fully analyze an SQL statement. In such cases, we recommend that you use the Detailed
Database Monitor.
46
API support for the Memory Resident Database Monitor
A set of APIs provides support for the Summary Monitor or Memory-based Database Monitor
that allow you to perform the activities listed in Table 3-1.
Table 3-1 External API description
API
Description
QQQSSDBM
This API starts the Memory-based Database Monitor. Database Monitor data is
collected in the threaded process but summarized at the job level.
QQQCSDBM
This API clears and frees the associated memory are of the SQL monitor.
QQQDSDBM
This API dumps the contents of the SQL monitor table. The API does not force a
clear operation (QQQCSDBM) of the memory. Data continues to be added to
memory until the QQQCSDBM or QQQESDBM API is called.
QQQESDBM
This API ends the memory-based SQL monitor.
QQQQSDBM
This API queries the status of the Database Monitor, and returns information about
the activity of the SQL and the original Database Monitor.
Figure 3-15 illustrates the different events in the Memory Resident Database Monitor life
cycle and the APIs associated with each event.
Start/Restart
(QQQSSDBM)
Start the SQL monitor
Continue
Pause
(QQQESDBM)
End the SQL monitor
(QQQDSDBM)
Dump the SQL monitor
Status
(QQQQSDBM)
Query status of the
database monitor
End
(QQQESDBM)
End the SQL monitor
(QQQDSDBM)
Dump the SQL monitor
(QQQCSDBM)
Clear SQL monitor memory
Figure 3-15 Memory Resident Database Monitor events and their APIs
For more information, search on Memory Resident Database Monitor external API
description in the V5R3 iSeries Information Center at:
http://publib.boulder.ibm.com/infocenter/iseries/v5r3/index.jsp
47
Notes:
򐂰 iSeries Navigator provides a graphical interface for these APIs, through the Summary
SQL Performance Monitor to administer the memory-based collection mode and to run
analytical dababase performance reports from the information collected.
򐂰 Unlike the Detailed Monitor, the Memory-based Database Monitor or Summary Monitor
outputs or dumps the collected information into 10 separate, categorized output tables.
To get the consolidated view of information collected for a single statement, you must
run join queries.
Memory Resident Database Monitor external table description
The Memory Resident Database Monitor uses its own set of tables instead of using the single
table with logical files that the STRDBMON monitor uses. The Memory Resident Database
Monitor tables closely match the suggested logical files of the STRDBMON monitor. The
tables are:
QAQQQRYI
QAQQTEXT
QAQQ3000
QAQQ3001
QAQQ3002
QAQQ3003
QAQQ3004
QAQQ3007
QAQQ3008
QAQQ3010
Query (SQL) information
SQL statement text
Table scan
Index used
Index created
Sort
Temporary table
Optimizer timeout/all indexes considered
Subquery
Host variable values
For more information about the definitions of these tables, search on Memory Resident
Database Monitor: DDS in the V5R3 iSeries Information Center:
3.7 Visual Explain
Visual Explain provides a graphical representation of the optimizer implementation of a query
request. The query request is broken down into individual components with icons that
represent each unique component. Visual Explain also includes information about the
database objects that are considered and chosen by the query optimizer. Visual Explain’s
detailed representation of the query implementation makes it easier to understand where the
greatest cost is incurred.
Visual Explain shows the job run environment details and the levels of database parallelism
that were used to process the query. It also shows the access plan in diagram form, which
allows you to zoom to any part of the diagram for further details as shown in Figure 3-16.
48
Figure 3-16 Visual Explain diagram
Visual Explain is a component of iSeries Navigator. From the Run SQL Script Center, you can
access Visual Explain directly, either from the menu or from the toolbar.
Note: The Run SQL Script interface always deletes ODPs to force full optimization.
Another way to access Visual Explain is through SQL Performance Monitor. SQL
Performance Monitor is used to create Database Monitor data and to analyze the monitor
data with predefined reports. For more information about the SQL Performance Monitor, refer
to Chapter 5, “Analyzing database performance data using iSeries Navigator” on page 93.
Visual Explain works with the monitor data that is collected by SQL Performance Monitor on
that system or by the Start Database Monitor (STRDBMON) command. Visual Explain can
also analyze Database Monitor data that is collected on other systems after data is restored
on the iSeries server.
For more information about the Visual Explain, refer to Chapter 8, “Analyzing database
performance data with Visual Explain” on page 197.
49
50
4
Chapter 4.
Gathering database SQL
performance data
After you identify a possible Structured Query Language (SQL) performance problem, you
must start gathering SQL performance data. In this chapter, we discuss the different tools that
are available to gather this data. Among these tools are the Detailed Database Monitor and
the Summary Database Monitor or Memory Resident Database Monitor.
We explain how to use iSeries Navigator to gather the Database Monitor data. We also
describe the table layout of the Database Monitor and the different record layouts of the
Database Monitor data.
51
4.1 Types of SQL Performance Monitors
Database Monitors have been part of the OS/400 operating system since V3R6. Database
Monitors and SQL Performance Monitors are used to gather information about queries run in
DB2 Universal Database for iSeries. This information can help you determine whether your
system and your queries are performing as they should, or whether they need fine tuning.
There are two types of SQL Performance Monitors:
򐂰 Detailed Database Monitors
򐂰 Summary Monitors (also known as Memory Resident Database Monitors)
The following sections describe the different ways that you can enable these two types of
monitors to gather database performance data.
4.2 Collecting monitor data
There are different ways to start a Database Monitor, as we explain in the following sections,
for:
򐂰
򐂰
򐂰
򐂰
Open Database Connectivity (ODBC) clients
Object Linking and Embedding (OLE) DB clients
Java Database Connectivity (JDBC) clients
Using an exit program
4.2.1 Starting a Detailed Database Monitor
The Start Database Monitor (STRDBMON) command starts a Detailed Database Monitor.
STRDBMON starts the collection of database performance statistics for a specified job or all
jobs on the system.
You can start the Detailed Database Monitor in a number of ways, all of which use the
STRDBMON command as the basis. You can issue the STRDBMON command from a
command line, a CL program, iSeries Navigator, and for ODBC, from within the data source
name. In this section, we look at all the options that are available to the database
administrator.
When you start a new monitor with the STRDBMON command, you must specify the file to
which the performance statistics are to be written. If the file or member does not exist, one is
created based on the QAQQDBMN file in library QSYS. If the file or member does exist, the
record format of the file is checked to see if it is the same.
To start the Detailed Database Monitor, from a command line, type one of the following
commands depending on whether you want to gather details for all the jobs on the system or
a particular job:
STRDBMON OUTFILE(MYLIB/DBMON) JOB(*ALL) TYPE(*DETAIL)
STRDBMON OUTFILE(MYLIB/DBMON) JOB(999999/SQLUSER/IBM) TYPE(*DETAIL)
The commands use the following parameters:
򐂰 OUTFILE
The file name for the results file is required, but the library name is optional. The file is
created if it does not exist, but is reused if it already exists.
52
򐂰 OUTMBR
This parameter defaults to the first member in the file. Specify the *ADD or *REPLACE
option. *REPLACE is the default option for this parameter.
򐂰 JOB
This parameter defaults to the job issuing the STRDBMON command. The user can
specify a single job on the system or specify *ALL for all jobs.
If the monitor is started on all jobs, any jobs waiting on job queues or any jobs started
during the monitoring period have statistics gathered from them after they begin. If the
monitor is started on a specific job, that job must be active in the server when the
command is issued.
Note: Each job in the server can be monitored concurrently by only two monitors:
򐂰 One started specifically on that job
򐂰 One started on all jobs on the server
When a job is monitored by two monitors and each monitor is logging to a different
output table, monitor rows are written to both logs for this job. If both monitors selected
the same output table, then the monitor rows are not duplicated in the output table.
򐂰 TYPE
This parameter allows the user to specify the type of data to be collected:
– *SUMMARY provides all necessary analysis data.
– *DETAIL collects the same data as *SUMMARY plus the 3019 row. This type of
collection causes a little more overhead on the system.
Tip: It is important to clarify that the STRDBMON command is a Detailed Database
Monitor regardless of which option is specified in the *DETAIL parameter. Specifying
*DETAIL is only useful for non-SQL queries, which are those queries that do not
generate a QQQ1000 row. For non-SQL queries, the only way to determine the number
of rows returned and the total time to return those rows is to collect detail rows.
Currently the only detail row is the 3019 row.
While the detail row contains valuable information, it creates a slight performance
degradation for each block of rows returned. Therefore you must closely monitor its use.
򐂰 FRCRCD
This parameter allows you to control the number of rows that are kept in the row buffer of
each job being monitored before forcing the rows to be written to the output table. The
default value is *CALC.
By specifying a force row write value of 1, FRCRCD(1) monitor rows are displayed in the
log as soon as they are created. FRCRCD(1) also ensures that the physical sequence of
the rows is most likely, but not guaranteed, to be in time sequence. However, FRCRCD(1)
causes the most negative performance impact on the jobs being monitored. By specifying
a larger number for the FRCRCD parameter, the performance impact of monitoring can be
reduced.
򐂰 COMMENT
This parameter allows you to add a meaningful description to the collection. It specifies the
description that is associated with the Database Monitor record whose ID is 3018.
Chapter 4. Gathering database SQL performance data
53
With STRDBMON, you can monitor SQL queries executed in DB2 Universal Database for
iSeries that come from different SQL interfaces such as ODBC, IBM Toolbox for Java, native
JDBC, or SQL call level interface (SQL CLI). Moreover, when you choose to collect detailed
information, you can also monitor queries that come from non-SQL interfaces such as
OPNQRYF and Query for iSeries.
4.2.2 Ending a Detailed Database Monitor
To end the collection of database performance data for a specified job or all jobs on the
system, you must use the End Database Monitor (ENDDBMON) command. The following
parameters are available for this command:
򐂰 JOB
You can specify the job name or *ALL to end only the monitor that was started with that
same job name. It is possible to end one monitor on a job and still have another one
collecting on that same job.
򐂰 COMMENT
This parameter enables you to describe the collection.
If an attempt is made to end the monitor on all jobs, then a STRDBMON command must have
been issued previously for all jobs. If a particular job is specified in this command, the job
must have the monitor started explicitly and specifically on that job.
For more information about the STRDBMON or ENDDBMON commands, visit the V5R3
iSeries Information Center at:
4.2.3 Enabling Database Monitors in ODBC clients
You can start a Detailed Database Monitor before a client initiates a connection to the server,
but using the STRDBMON command to monitor the client job. Other options are available for
you to start a Database Monitor. The two ways to start a Detailed Database Monitor in ODBC
clients are:
򐂰 Enable the Database Monitor option in the data source name
򐂰 Use an ODBC connection keyword to start the Database Monitor
ODBC data source name
To start a Detailed Database Monitor using the data source name (DSN):
1. In Microsoft® Windows XP, click Start → Programs → IBM iSeries Access for Windows
and then select ODBC administration.
2. From the ODBC Data Source Administrator window (Figure 4-1), select the desired data
source name and click the Configure button.
54
Figure 4-1 ODBC Data Source Administrator window
3. In the iSeries Access for Windows ODBC setup window (Figure 4-2) that opens, click the
Diagnostic tab and then select the Enable Database Monitor option. Then click OK.
Figure 4-2 iSeries Access for Windows ODBC Setup window
The Enable Database Monitor option causes the ODBC driver to issue a call to STRDBMON
for the job connecting to this data source name. The output file is created in the QUSRSYS
55
library starting with the prefix QODB and ending with the job number, for example
QUSRSYS/QODB337344.
The ODBC driver attempts to end the Database Monitor when the application disconnects. If
the application ends abnormally without issuing a disconnect, the Database Monitor might
continue to run. In this case, you must manually end the monitor by entering the ENDDBMON
command and specifying the job number being traced (or *ALL if no other Database Monitors
are active).
ODBC connection keywords
One potential problem with redistributing an application that uses ODBC is that a data source
might need to be created on each user’s PC. Data sources are normally created using ODBC
C APIs. However, this interface might be difficult to use for some programming languages.
An alternative for this potential problem is for a client to connect to the server without using an
ODBC data source and to use connection keywords instead. The iSeries Access ODBC
driver has many connection string keywords that can be used to change the behavior of the
ODBC connection. These same keywords and their values are also stored when an ODBC
data source is setup.
Tip: When an ODBC application makes a connection, any keywords specified in the
connection string override the values that are specified in the ODBC data source.
The connection keyword to enable the Database Monitor is the TRACE keyword. Currently
the TRACE keyword supports the following options:
򐂰
򐂰
򐂰
򐂰
򐂰
0 = No tracing
1 = Enable internal driver tracing
2 = Enable Database Monitor
4 = Enable the Start Debug (STRDBG) command
16 = Enable job trace
To specify multiple trace options, add together the values for the options that you want. For
example, if you want to activate the Database Monitor (2) and the STRDBG command (4) on
the server, then the value that you specify is 6 (2+ 4=6).
Important: Use these options only when debugging problems because they can adversely
affect performance.
The part of a Microsoft Excel® macro in Example 4-1 uses a data source name and the
TRACE connection keyword to enable the Database Monitor and to enable the STRDBG
command. Because this example uses both a data source name and a connection keyword,
the value in the TRACE keyword overrides the values specified in the data source name.
Example 4-1 Microsoft Excel macro using a data source name and TRACE connection keyword
'ODBC connection.
'The system is specified in the B1 cell in the Settings worksheet
ConnectStr = connectStr & "DSN=QDSN_TPLXE2;System=" &
Worksheets("Settings").Range("B1").Value & ";TRACE=6"
CnMYAS400.Open connectStr
The TRACE keyword, including the value to Enable Database Monitor, causes the ODBC
driver to issue a call to STRDBMON for the current job. The output file is created in the
56
QUSRSYS library, starting with the prefix QODB, and ending with the job number, for
example QUSRSYS/QODB337344.
4.2.4 Enabling Database Monitors in OLE DB clients
There are two ways you can start a Detailed Database Monitor in OLE DB clients:
򐂰 Using an OLE DB connection property
򐂰 Using an OLE DB connection keyword
OLE DB connection properties
A set of custom properties (IBMDA400, IBMDARLA, and IBMDASQL) is available for the OLE
DB providers shipped with iSeries Access for Windows. The trace property (available in
V5R3) is used to enable diagnostic traces when troubleshooting errors. It is an integer
property, and several numeric constants are defined for various trace options.
To determine the value this property should contain, select the desired trace options and add
the constant values. The resulting number is the value that should be specified. The
constants are:
򐂰
򐂰
򐂰
򐂰
򐂰
0 = No trace options (the default value)
2 = Enable the STRDBG command
4 = Print Job Log at disconnect
8 = Enable Job trace via the Start Trace (STRTRC) command
Example 4-2 shows how to enable the Database Monitor using Visual Basic®.
Example 4-2 Enabling the Database Monitor using Visual Basic
Dim cnAS400 as ADODB.Connection
Dim strJobName as String
Set cnAS400 = New ADODB.Connection
'Set the provider to Client Access
cnAS400.Provider = "IBMDA400"
'Set custom properties.
cnAS400.Properties("Block Fetch") = True
cnAS400.Properties("Catalog Library List") = "LIBRARY1, LIBRARY2"
cnAS400.Properties("Convert Date Time To Char") = "FALSE"
cnAS400.Properties("Default Collection") = "MYLIB"
cnAS400.Properties("Force Translate") = 0
cnAS400.Properties("Cursor Sensitivity") = 0
cnAS400.Properties("Data Compression") = True
cnAS400.Properties("Hex Parser Option") = 0
cnAS400.Properties("Initial Catalog") = "*SYSBAS"
cnAS400.Properties("Maximum Decimal Precision") = 31
cnAS400.Properties("Maximum Decimal Scale") = 31
cnAS400.Properties("Minimum Divide Scale") = 0
cnAS400.Properties("Query Options File Library") = "QUSRSYS"
cnAS400.Properties("SSL") = "FALSE"
cnAS400.Properties("Trace") = 1 'Enable Database Monitor
'Open the connection
cnAS400.Open "Data Source=MySystem;", "USERID", "PWD"
strJobName = cnAS400.Properties("Job Name")
57
OLE DB connection keywords
In addition to using the trace property to enable Database Monitor in OLE DB, you can use
the Trace connection keyword. In Example 4-3, we illustrate an Excel macro that uses the
Trace connection keyword to enable Database Monitor at connection time.
Example 4-3 Excel macro using the Trace connection keyword
'OLE DB Connection
connectStr = connectStr & "provider=IBMDA400;data source=" &
Worksheets("Settings").Range("B1").Value & ";TRACE=1"
cnMYAS400.Open connectStr
The trace property or the trace connection keyword, including the value to Enable Database
Monitor, causes the OLE DB Provider to issue a call to STRDBMON for the current job. The
output file is created in the QUSRSYS library, starting with the prefix QODB and ending with
the job number, for example QUSRSYS/QODB337344.
4.2.5 Enabling Database Monitors in JDBC clients
The JDBC driver shipped with the Developer Kit for Java (commonly known as the native
JDBC driver) and the IBM Toolbox for Java JDBC driver support a server trace connection
property. Among other options, it includes an option to start a Detailed Database Monitor.
Important: These tracing options work correctly only if you are using an SQL naming
convention. They do not work if the JDBC connection is created using a system naming
convention. A future version of the drivers will address this issue with these options.
In Example 4-4, the Java code uses a properties object to enable tracing. The example uses
native JDBC and the IBM Toolbox for Java JDBC concurrently.
Example 4-4 Java code using JDBC
// Register both drivers.
try {
Class.forName("com.ibm.db2.jdbc.app.DB2Driver");
Class.forName("com.ibm.as400.access.AS400JDBCDriver");
} catch (ClassNotFoundException cnf) {
System.out.println("ERROR: One of the JDBC drivers did not load.");
System.exit(0);
}
Connection conn1, conn2;
Properties props = new Properties();
props.setProperty("user", "MYUSER");
props.setProperty("password", "MYPASSWORD");
props.setProperty("server trace", "n");
try {
// Obtain a connection with each driver.
conn1 = DriverManager.getConnection("jdbc:db2://localhost", props);
conn2 = DriverManager.getConnection("jdbc:as400://localhost", props);
conn1.close();
conn2.close();
} catch (SQLException e) {
System.out.println("ERROR: " + e.getMessage());
}
58
In this example, n is a numeric bitmap of selected trace options. The options that can be
specified include:
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
1 = Client trace
4 = Debug job log
8 = Save job log
16 = Job trace
32 = Save SQL output
The Database Monitor option causes the drivers to issue a call to STRDBMON for the current
job. The output file is created in the QUSRSYS library, starting with the prefix QJT for the
toolbox driver and the prefix QSQL for the native driver, and ending with the job number for
example, QUSRSYS/QJT744340 or QUSRSYS/QSQL744340.
4.2.6 Enabling Database Monitors using an exit program
An exit program provides another way to start a Detailed Database Monitor for clients using
the QZDASOINIT, QZDASSINIT, or QZDAINIT prestart jobs, such as ODBC, OLE DB, or IBM
Toolbox for Java clients. The database server has five different exit points defined.
QIBM_QZDA_INIT is one of those exit points and is called when the prestart jobs are started.
By using the Work with Registration Info (WRKREGINF) command, you can add or remove
the exit program, which has a STRDBMON command, to the exit point. You must end and
restart the prestart jobs using the Start Prestart Jobs (STRPJ) and End Prestart Jobs
(ENDPJ) commands for the change to take effect. While we do not discuss this option further,
we strongly recommend that you use one of the options previously described. To learn more
about exit programs, search on register exit programs in the V5R3 iSeries Information
Center at:
4.3 Collecting monitor data using iSeries Navigator
In this section, we explain how to start the two types of SQL Performance Monitors using
iSeries Navigator.
4.3.1 Starting a Memory Resident or Summary Database Monitor
To start the Memory Resident or Summary Database Monitor within iSeries Navigator,
right-click SQL Performance Monitors and select New → Summary as shown in Figure 4-3.
As the name implies, this monitor resides in memory and only retains a summary of the data
collected. When the monitor is paused or ended, the data is written to hard disk and can then
be analyzed. Because the monitor stores its information in memory, the performance impact
to your system is minimized.
59
Figure 4-3 Starting a Summary Database Monitor using iSeries Navigator
In the New Summary SQL Performance Monitor window (Figure 4-4) that opens, you can
name the Summary Monitor and select the jobs that you want to monitor and the specific data
that you require the monitor to collect.
Figure 4-4 General tab
60
Figure 4-4 shows the General tab on which you provide the following information:
򐂰 Name (of the summary monitor)
In this field, you specify the name of the monitor. This name must be unique in the schema
in which it is created, can be up to 30 characters long, and exists only in iSeries Navigator.
This is a required field.
򐂰 Schema (where the monitor will reside on the iSeries server)
In this field, you specify the schema where the monitor is saved. The system automatically
folds all lowercase characters to uppercase. You can change this by placing the schema
name with special characters within double quotation marks. This field is required to start
the monitor.
򐂰 Storage (amount that you will allow the monitor to you)
This field specifies the size limit of the monitor. The default value is not the limit. If the limit
is reached, the monitor continues collecting the data at the beginning of the table,
overwriting previous data.
The Monitored Jobs tab, shown in Figure 4-5, allows you to specify which jobs to monitor. You
can choose which jobs you want to monitor or choose to monitor all jobs.
Figure 4-5 Monitored Jobs tab
You can have multiple instances of monitors running on you system at one time. However,
only one monitor instance can monitor all jobs. Additionally, you cannot have two monitors
monitoring the same job. When collecting information for all jobs, the monitor collects on
previously started jobs or new jobs started after the monitor is created. You can edit this list by
selecting and removing jobs from the Selected jobs list.
61
In the Data to Collect tab (Figure 4-6), select the options for the kind of data that you want to
collect.
Figure 4-6 Options for Summary Monitor
The options on the Data to Collect tab collect the following information:
򐂰 Summary data: Contains resource and other general information about monitored jobs;
option available only on this page.
򐂰 Statement text: Contains the SQL text that monitored jobs call
򐂰 Host variable use: Contains the values of host variables that monitored jobs use
򐂰 Data sorts: Contains details of data sorts that monitored jobs perform
򐂰 Table scan: Contains the table scan data for the monitored jobs
򐂰 Index creation: Contains details of the creation of indexes by monitored jobs
򐂰 Indexes used: Contains details of how indexes are used by monitored jobs
򐂰 Optimizer timeout/access paths considered: Contains details about any occurrences of
timeouts of monitored jobs
򐂰 Subselect processing: Contains information about each subselect in an SQL statement
򐂰 Temporary file use: Contains details about temporary files that monitored jobs created
Finally, you click OK to start the Summary Database Monitor. The new monitor should be
listed in the right pane of the iSeries Navigator window with the type Summary and a status of
Started.
You can pause or end a Summary Monitor, or you can continue a monitor that was previously
paused. However, you cannot continue a monitor that was previously ended. When the
monitor is paused or ended, the monitored data that resides in memory is written into several
database tables that can be queried and analyzed using the predefined reports that come
with the tool.
62
After the data is collected, you can look at the properties of the monitor. In the SQL
Performance Monitor window, you right-click the monitor in question and click Properties.
Figure 4-7 shows the properties of a collected Summary Monitor. Notice the File column, in
which the files or tables reside in the schema that was specified when the monitor was
started. On this page, you also see the file names and the collection period of the data for
which the monitor ran. If you are viewing an imported monitor, the statement “Information
Not Available” is displayed in the Collection period field.
Figure 4-7 Summary Saved Data tab
External table description
The following externally described file definitions apply to those shown in Figure 4-7. For your
reference, each file includes the file name for clarification.
Tip: The file names as they are shown in Figure 4-7 are specific to this iSeries server.
They will change with each collection.
򐂰 QAQQQRYI Query (SQL) information (QPM0000158)
Displays the file that contains the summary data
򐂰 QAQQTEXT SQL statement text (QPM0000159)
Displays the file that contains the statement text
򐂰 QAQQ3000 Table scan (QPM0000140)
Displays the file that contains the table scan
򐂰 QAQQ3001 Index used (QPM0000141)
Displays the file that contains index use
63
򐂰 QAQQ3002 Index created (QPM0000142)
Displays the file that contains index creation
򐂰 QAQQ3003 Sort (QPM0000143)
Displays the file that contains the data sorts
򐂰 QAQQ3004 Temporary table (QPM0000144)
򐂰 Displays the file that contains temporary file use
򐂰 QAQQ3007 Optimizer timeout/all indexes considered (QPM0000147)
Displays the file that contains optimizer or access paths
򐂰 QAQQ3008 Subquery (QPM0000148)
Displays the file that contains information about each subselect in an SQL statement
򐂰 QAQQ3010 Host variable values (QPM00000150)
Displays the file that contains the host variable use
Note: To see the data description specification (DDS) for the file, see Chapter 12 in DB2
Universal Database for iSeries Database Performance and Query Optimization, which is
available on the Web at:
http://publib.boulder.ibm.com/infocenter/iseries/v5r3/topic/rzajq/rzajqmst.pdf
When to use the Summary Database Monitor
The Memory Resident or Summary Database Monitor, as already discussed, has advantages
over the Detailed Database Monitor created when using the STRDBMON command. The
Memory Resident Database Monitor can be more useful than the Detailed Database Monitor
in the following situations:
򐂰 When you are not sure what is causing the performance issue that you might be seeing
You can collect a lot of data over the entire system, in a short amount of time, using the
Detailed Monitor. It is better to collect a Summary Monitor over the entire system and
analyze that.
򐂰 When you want to monitor the system over a period of time to compare, for example, the
results of one week against another
This helps with a proactive approach to SQL performance and with trends in the way the
database is performing.
򐂰 When the size of the data collected for a Summary Monitor started against the whole
system is still significantly smaller
This can help the customer provide IBM Support with initial problem determination data.
Tip: IBM Support might request a Detailed Monitor for more detailed analysis of the issues.
Also the space used by the Memory Resident Database Monitor on the system is
significantly less. Two monitors, side-by-side over the same SQL functions, collected only
around 5% of the full Detailed Database Monitor.
64
Data conversion problems
You can use the Memory Resident Database Monitor to analyze possible problems with data
conversion issues. Collect the Database Monitor data over the system and then use the code
shown in Example 4-5 to evaluate the data.
Example 4-5 SQL to run Memory Resident Database Monitor to show data conversion issues
select count(*) as Conversions, a.qqdacv, b.qqsttx
from qgpl.qpm0000158 a, qgpl.qpm0000159 b
where a.qqkey = b.qqkey and a.qqdacv between '1' and '9'
group by a.qqdacv, b.qqsttx
order by 2 asc
A single data conversion operation is inexpensive, but repeated thousands or millions of
times, it can add up. In some cases, it is a simple task to change one of the attributes so that
a faster direct map can be performed. In other cases, the conversion is necessary because
no exact matching data type is available. One of the advantages of the Memory Resident
Database Monitor, as previously discussed, is the affect that the monitor has on the system.
In this example, it is easy to see whether you have data conversion issues.
4.3.2 Starting a Detailed Database Monitor
Using the detailed SQL Performance Monitor in iSeries Navigator is equivalent to running the
STRDBMON command with *DETAIL in the TYPE parameter, which is explained in 3.6.1,
“Detailed Monitor” on page 44.
To start a detailed SQL Performance Monitor on the iSeries Navigator window, right-click SQL
Performance Monitors and select New → Detailed as shown in Figure 4-8.
Figure 4-8 Starting the Detailed Database Monitor on iSeries Navigator
65
In the New Detailed SQL Performance Monitor window (Figure 4-9) that opens, under the
General tab, specify the name of the monitor and the schema to be saved.
Figure 4-9 General tab of the Detailed SQL Performance Monitor window
In the Monitored Jobs tab, select the jobs that you are interested in monitoring. You can
specify all jobs or select from the list of jobs as shown in Figure 4-10.
Figure 4-10 Monitored Jobs in the Detailed SQL Performance Monitor window
66
Selecting All (for all jobs) might cause a significant overhead on the system, particularly if it is
a detailed SQL Performance Monitor.
Finally, click OK and the Detailed SQL Performance Monitor begins. You should see the new
monitor listed in the right pane of iSeries Navigator, with a Type of Detailed and a Status of
Started, as shown in Figure 4-11.
Figure 4-11 SQL Performance Monitor status pane
The monitoring of that job will continue for as long as it is on the system or until you end the
monitor. To end a monitor, right-click the monitor and click the End option (see Figure 4-12).
Then you can use this data to further analyze your query with the tools described in this
redbook.
Unlike the Summary SQL Performance Monitor, you cannot pause a Detailed SQL
Performance Monitor. However, you do not need to end the monitor to start analyzing the
results. You can use the predefined reports to analyze the data as described in Chapter 5,
“Analyzing database performance data using iSeries Navigator” on page 93. Or you can write
your own queries to query the monitored data as described in Chapter 6, “Querying the
performance data of the Database Monitor” on page 133.
67
Figure 4-12 Ending an SQL Performance Monitor
Alternatively, you can start a Detailed SQL Performance Monitor from a Run SQL Script
window in iSeries Navigator. In this case, the monitor is started only for the job associated
with the Run SQL Script window. You can perform other functions from a Run SQL Script
window, including to end, analyze, and delete the monitor, as well as to look at the monitor’s
properties as shown in Figure 4-13.
Figure 4-13 Start SQL Performance Monitor from the Run SQL Scripts window
68
When you end an SQL Performance Monitor, you cannot start it again, but you can continue
with an SQL Performance Monitor (Summary) that was previously paused.
4.3.3 Importing Database Monitors into iSeries Navigator
In the Import SQL Performance Monitor Files window, you can incorporate data for SQL
Performance Monitors that were started without using iSeries Navigator, or data that was
collected on another system. To access this window, from the main iSeries Navigator window,
right-click SQL Performance Monitors and select Import.
In the Import SQL Performance Monitor Files window (Figure 4-14), if you are importing data
collected by a Memory Resident Database Monitor, for Type of monitor, select Summary. You
must only specify the name of any one of the data files. Click OK to import the monitor.
Figure 4-14 Import SQL Performance Monitor window
69
You should see the imported monitor in the list of monitors in the main iSeries Navigator
window, with the status of Imported, as shown in Figure 4-15.
Figure 4-15 Imported monitors
70
4.4 SQL Performance Monitors properties
At any time, you can display the properties of a monitor in iSeries Navigator. Right-click the
monitor for which you want to display the properties and select Properties as shown in
Figure 4-16.
Figure 4-16 Selecting to view the properties of an SQL Performance Monitor
71
In the Properties window, click General tab (Figure 4-17) to view general information such as
the monitor name, the schema to be saved, and the storage limit for the monitor. On this
page, you also see the type of monitor (Summary or Detailed), the status, and the user profile
that created the monitor.
Figure 4-17 General tab of the Properties window
72
Click the Monitored Jobs tab (Figure 4-18), in the Properties window, to see the list of
monitored jobs.
Figure 4-18 Monitored Jobs tab of the Properties window
Click the Saved Data tab (Figure 4-19) to see information based on the type of monitor:
򐂰 Collection period: Specifies the monitoring period for which you want to view data
If an imported monitor is viewed, you see the statement “Information Not Available” in
the Collection period box. The monitoring period can only be changed on the Properties
window of the Summary Monitor.
򐂰 Schema: The schema where the monitor tables reside
򐂰 Monitor data selected: For a detailed SQL Performance Monitor, this section shows the
name of the tables, for each one of the options selected when the monitor was started.
The name of the tables start with the prefix QPM, followed by a sequential number. The
tables have the same format as the Memory Resident Database Monitor tables (see
“External table description” on page 63).
73
Figure 4-19 Saved Data tab of the Properties window for a Summary Monitor
For a detailed SQL Performance Monitor, you see the name of the table where the data is
stored (Figure 4-20). The name of the tables starts with the prefix QZG, followed by a number
assigned by the system. The table has the same format as when created with STRDBMON.
Figure 4-20 Saved Data tab for a Detailed Database Monitor
74
For monitors that were imported into iSeries Navigator, you see the name of the original
monitor table that was used as the source table when the import took place (see Figure 4-21).
Figure 4-21 Saved Data tab of an Imported monitor
4.4.1 Considerations for the SQL Performance Monitors in iSeries Navigator
SQL Performance Monitors in iSeries Navigator only display the monitors that were created
using iSeries Navigator (Summary and Detailed) or monitors that were imported. You can
import monitor files that reside in different libraries on your system and that were created
using other methods. You can use iSeries Navigator as the central application to work with all
your monitors on the system. Moreover, you can take advantage of predefined reports to
analyze the monitor data even for monitors that were imported into iSeries Navigator.
4.5 Summary or Detailed Database Monitor
There are circumstances in which it is better to collect a Summary SQL Performance Monitor
than to collect a Detailed SQL Performance Monitor or vice versa. In this section, we explain
when one is more useful than the other.
The Memory Resident Database Monitor has advantages over the Detailed Database Monitor
created when using the STRDBMON command. The Memory Resident Database Monitor
can be more useful than the Detailed Database Monitor in the following situations among
others:
򐂰 When you are not sure what is causing the performance issue that you might be seeing
You can collect a lot of data over the entire system in a short period of time using the
Detailed Database Monitor, which causes a big overhead. It is better to collect a Summary
Monitor over the entire system and analyze that.
75
򐂰 When you want to monitor the system over a set period of time to compare the results of
one week against another, for example
This helps with a proactive approach to SQL performance and with trends in the way that
the database is performing.
򐂰 When the size of the data collected for a Summary Monitor started against the whole
system is still significantly smaller than the data collected for a Detailed Monitor
This can help the customer provide IBM Support with initial problem determination data.
A Detailed Database Monitor can be more useful when:
򐂰 You need more detailed information about your queries and not just summary data.
򐂰 You need to analyze non-SQL queries and see information through the 3019 detailed
record.
򐂰 You want to see the implementation of your SQL queries and use tools, such as Visual
Explain in iSeries Navigator, to explain the queries.
򐂰 You want to check for the generation of SQLCODE and SQLSTATE errors on your queries.
Important: The data collection in a Database Monitor is done inline within the job instead
of a separate job.
Optimization records and data do not appear for queries that are already in reusable open
data path (ODP) mode when the monitor is started. To ensure the capture of this data for a
batch job, start the monitor before the job starts and collect it over the entire length of the
job or as much as needed.
76
4.6 The Database Monitor record types
The Detailed Database Monitor collects different data and stores records in a single table in
the order of occurrence. Within the Database Monitor table, each record contains a record
type column. The Database Monitor uses the QQRID column to describe the type of
information gathered in the particular record. Figure 4-22 shows the Database Monitor record
types that are most often used for performance analysis.
Figure 4-22 Detailed and Extended Detailed Results and their main source of information
Note: In the sections that follow, we identify the columns for each record type. The term
“column” is also known as field.
4.6.1 Database Monitor record types
In this section, we list the Database Monitor record types that are most often used for
performance analysis, as well as other record types. We also list the Global Database Monitor
data columns and other columns that identify the tables or indexes used.
Record types most often used (QQRID value)
The following record types are most often used:
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
1000 Record: SQL statement summary (see page 80)
3000 Record: Arrival sequence (table scan; see page 82)
3001 Record: Using existing index (see page 83)
3002 Record: Temporary index created (page 85)
3003 Record: Query sort (see page 86)
3004 Record: Temporary file (see page 87)
3006 Record: Access plan rebuild (see page 87)
77
򐂰 3007 Record: Summary optimization data (also known as index optimization data; see
page 88)
򐂰 3010 Record: Host variable and ODP implementation (see page 89)
򐂰 3021 Record: Bitmap created
򐂰 3022 Record: Bitmap merge
򐂰 3023 Record: Temporal hash table created
򐂰 3026 Record: Union merge
򐂰 3027 Record: Subquery merge
򐂰 3028 Record: Grouping
򐂰 3029 Record: Index ordering
Database Monitor record types: Other record types
The following record types are in the Database Monitor table but are not used in all the
records:
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
3005 Record: Table locked
3008 Record: Subquery processing
3014 Record: Generic query information
3018 Record: STRDBMON and ENDDBMON data
3019 Record: Retrieved detail (only with *DETAIL)
3025 Record: DISTINCT processing
3030 Record: Query step processing
5002 Record: SQL request executed by SQL Query Engine (SQE)
Record types 3000 to 3008, 3014, and 3021 to 3029 occur during a full open and can be
referred to as optimization records. Optimization records are much like debug messages.
These records are necessary to determine the access plan for any given query in the
Database Monitor data.
Global Database Monitor data columns
The following data columns are common to all record types:
򐂰 QQJOB (job name)
򐂰 QQUSER (job user name)
򐂰 QQJNUM (job number): The job number is useful when multiple jobs are collected in one
DB Monitor file.
򐂰 QQJFLD (join field): This column contains information that uniquely identifies a job and
includes job name, job user name, and job number.
򐂰 QQTIME (time at which the record was created): The time record can be useful when
trying to determine which queries were running in a given time period.
򐂰 QQUCNT (unique number given for each query within a job): QQUCNT links together all
Database Monitor records associated with all instances of a unique query within a job. The
QQUCNT value assigned at full open time stays constant for all subsequent instances of
that query. Non-ODP SQL operations (prepare, describe, commit) have QQUCNT = 0 and
cannot be linked to a query. But the QQ1000 column in the prepare or describe 1000
record contains the prepared SQL text.
This data column is not set for optimization records.
򐂰 QQI5 (refresh counter): This record specifies the instance number for a unique query. It is
used in conjunction with the QQUCNT value to look at a specific instance of a query and is
only valid on 3010 and 1000 SQL summary records.
Non-ODP 1000 records (commit, prepare, and so on) have QQI5 = 0.
򐂰 QQRID (record identifier): This column identifies the type of record.
78
򐂰 QVC102: This column refers to the CURRENT job user name.
򐂰 QQ19: (thread identifier): This column might be useful for multithreaded applications.
Figure 4-23 shows some of the Global Database Monitor columns.
Figure 4-23 Global Database Monitor columns
Tip: If you are going to use the Run SQL Script in iSeries Navigator, change the JDBC
setup to force translation for CCSID 65535, because the QQJFLD has been defined as
FOR BIT DATA. Otherwise this column is shown in hexadecimal.
Reset the default value to use Visual Explain; otherwise it will fail.
Other columns that identify tables or indexes used
The following columns can also identify the tables or indexes that are used:
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
QQTLN: Library of the table queried
QQTFN: Name of the table queried
QQPTLN: Base library
QQPTFN: Base table
QQILNM: Library of the index used
QQIFNM: Name of the index used; *N when it is a temporary index
QVQTBL: Queried table long name
QVQLIB: Queried library long name
QVPTBL: Base table long name
QVPLIB: Base table library long name
QVINAM: Index used long name
QVILIB: Index library long name
79
How the data is organized in the Database Monitor table
The first occurrence of a unique query within the job always results in full open. A “unique”
query is one that requires a new ODP. SQL has determined that there is no existing ODP that
can be used.
The presence of optimization records indicates a full open for an Open, Select into, Update,
Delete, or Insert operation. Optimization records are immediately followed by SQL summary
records (QQRID=1000) for that operation.
Subsequent occurrences of this query within the same job either run in reusable ODP or
nonreusable ODP mode. Nonreusable mode is indicated by the presence of optimization
records each time a particular query is run (full open). Reusable ODP mode is indicated by
3010 and 1000 records each time the given query is run (no optimization records or full
open).
Linking query instances in the Database Monitor data
The data in the Database Monitor file is arranged chronologically. This organization can make
it difficult to find all instances of a unique query. Use the QQJFLD, QUCNT, and QQI5
columns to view specific query instances. Be aware of the following situations:
򐂰 QUCNT is assigned during a full open and is constant for all subsequent instances of a
query.
򐂰 Non-ODP SQL operations (prepare, describe, commit) have QQUCNT = 0 and, therefore,
cannot be linked to a query. If this is the case, you can use the QQ1000 column, which in
the Prepare or Describe operation, contains the prepared SQL text.
򐂰 Non-ODP 1000 records (commit, prepare, and so on) have QQI5 = 0.
A full open occurs when there is an SQL update, insert, delete, or open operation and the
QQI5 record is 0.
4.6.2 The 1000 Record: SQL statement summary
The 1000 record is the basic record type for any SQL query analysis. One record exists for
each SQL operation (open, update, close, commit, and so on). The following columns are
those that are most commonly used:
򐂰 QQ1000: Prepared text of SQL statement literals in the original SQL text might be
replaced by parameter markers in prepared text if SQL was able to convert them during
preparation (desired). For the original SQL text, use literal values from the matching 3010
record in the place of parameter markers or obtain the text from the step mode file using
the QQSTIM time stamp from this record.
򐂰 QQC21: This column indicates the type of SQL operation (OP, FE, CL, UP, IN, DL, and so
on). A value of MT in this column indicates the continuation of a record for SQL statements
that exceed 1000 characters. A value of FE indicates a Fetch summary record and not the
actual number of fetch operations.
The ODP-related operation types are OP, IN, UP, DL, SI, and SK.
򐂰 QQI2: Number of rows updated, inserted, or deleted
򐂰 QQI3: Number of rows fetched (only on FE rows)
This column indicates the actual number of rows fetched, and not the number of fetched
attempts.
80
򐂰 QQI6: Elapsed time for this operation in microseconds
The time to fetch all rows might not be included in with Open and Select operations; you
must look at the time on Fetch operation rows.
Access plan information
The following columns indicate information about the access plan:
򐂰 QQC103 and QQC104: Package or program name and library
򐂰 QVC18: Dynamic SQL statement type
A value of E represents “extended dynamic”, a value of S represents “system wide”
statement cache, and a value of L represents a prepared statement.
򐂰 8 and QVC22: Access plan rebuild code and subcode
This subcode is useful for IBM debug purposes.
򐂰 QVC24: Access plan save status
A value of Ax means that the access plan could not be saved. Values of Bx or a blank
mean that the access plan was saved successfully.
ODP information
The following columns provide information about the ODP:
򐂰
򐂰
򐂰
򐂰
򐂰
QQI5: Query instance counter, where a value of 0 means that a full open occurred
QQC15: Hard close reason code (for an HC operation type)
QVC12: Pseudo open indicator
QVC13: Pseudo close indicator
QQC181 and QQC182: Cursor and statement name
Example 4-6 shows a query with some of the most commonly used columns in the 1000
record.
Example 4-6 Common columns in the 1000 record
SELECT qqucnt AS "QQUCNT Unique Counter"
,qqc21 AS "QQC21 Statement Operation"
,qqetim - qqstim AS "Elapsed Time"
,qq1000 AS "QQ1000 Text"
,qqi3 AS "QQI3 Fetched Rows"
,qqc16 AS "QQC16 Data Conv."
,qvc11 AS "QVC11 ALWCPYDTA"
,QVC41 AS "QVC41 Cmt Control Lvl"
,rrn(a)
FROM vldbmon a
WHERE qqrid = 1000
ORDER BY rrn(a)
OPTIMIZE FOR ALL ROWS;
81
Figure 4-24 shows the result of the query in Example 4-6. The first row shows that the Open
operation took 603 seconds and only 0.09 seconds to fetch 125 rows. The query ran with
ALWCPYDTA *OPTIMIZE and with commitment control set to *NONE. Also, there were no
conversion problems when the query was run.
Figure 4-24 Common columns in the 1000 record
4.6.3 The 3000 Record: Arrival sequence (table scan)
The 3000 record points outs queries in which the entire table is scanned. A table scan is
generally acceptable in cases where a large portion of the table will be selected or the table
has a small number of rows.
Table information
Table information is provided by the following columns:
򐂰 QVQTBL and QVQLIB: Table name and schema name respectively
򐂰 QQPTFN and QQPTLN: Short system table name and library name respectively
򐂰 QQTOTR: Number of rows in table
Query optimization details
The following columns provide details about query optimization:
򐂰 QQRCOD: Reason code, why table scan chosen
򐂰 QQIDXA: Index advised (Y or N)
If the value is N, QQI2 and QQIDXD will not contain data.
򐂰 QQI2: Number of primary (key positioning) keys in QQIDXD column
򐂰 QQIDXD: Suggested keys for index (selection only)
This column can contain both primary and secondary keys. Starting from the left, QQI2
indicates the number of keys that are considered primary.
82
The query shown in Example 4-7 illustrates some of the most commonly used columns in the
3000 record.
WITH xx AS (SELECT * FROM vldbmon WHERE qqrid = 3000),
yy AS
(SELECT qq1000 AS qqsttx
,qqjfld ,qqucnt ,qqc21 as qqop ,qqi4 as qqtt FROM vldbmon
WHERE qqrid = 1000 AND qqc21 <> 'MT'
AND (qvc1c = 'Y' OR (qqc21 IN('DL', 'UP')
AND qqc181 <= ' ') OR qqc21 IN ('IN', 'IC', 'SK', 'SI') OR qqc21 LIKE 'O%'))
SELECT qqop as "Operation" ,qqtt as "Total time" ,qqptln as "Library" ,qqptfn as "Table"
,qqtotr as "Rows" ,qqrcod as "Reason" ,qqidxa as "Index Advised" ,qqi2 as "Primary
Keys" ,qqidxd as "Suggested keys" ,qqsttx as "Statement" FROM xx a LEFT JOIN yy b ON
a.qqjfld = b.qqjfld AND a.qqucnt = b.qqucnt ORDER BY qqidxa DESC;
Figure 4-25 shows the result of the query in Example 4-7. The information from the 3000
record is joined with information from the 1000 record, to determine which queries did a table
scan. Therefore, in a case where an index is not recommended, we can still look at the
selection in the SQL text to see if a good index can be created.
4.6.4 The 3001 Record: Using an existing index
The 3001 record shows the index that will be used to access the table and why it was chosen.
If the index was chosen for a join operation, additional information is given to help determine
how the table “fits” in the join. The order of the 3001 records indicates the join order chosen
by the optimizer.
Index and table information
The following columns provide index and table information:
򐂰 QVINAM and QVILIB: Name of the chosen index and library
򐂰 QVQTBL and QVQLIB: Name of the associated table and library
Query optimization details
The following columns indicate details about query optimization:
򐂰 QQRCOD: Reason the index was selected
I1
I2
I3
I4
I5
Selection only
Ordering or grouping
Selection and ordering or grouping
Nested loop join
Record selection using bitmap
83
򐂰
򐂰
򐂰
򐂰
򐂰
QQIDXA: Index advised (Y or N)
QQI2: Number of primary (key positioning) keys in QQIDXD column
QQIDXD: Suggested keys for index (selection only)
QVC14: Index only access indicator (Y or N)
QQIA: Index page size
Note: A combination of the 3000 and 3001 records for a table indicates bitmap processing.
Example 4-8 shows a query with some of the most commonly used column in the 3001
record.
SELECT qqilnm as "Library"
,qqidxa as "Index Advised"
,qvc14 as "Index Only"
,qqia as "Index page size"
FROM vldbmon a
WHERE qqrid = 3001
order by qqidxa desc;
,qqifnm as "Table" ,qqtotr as "Rows" ,qqrcod as "Reason"
,qqi2 as "Primary Keys" ,qqidxd as "Suggested keys"
Figure 4-26 shows the result of the query in Example 4-8. It shows the following information:
򐂰 ITEM_00003 was used for ordering and grouping (reason I2). However, the optimizer is
advising an index with two primary keys most likely for selection.
򐂰 Some indexes are temporary and were created to satisfy the join in the query (reason I4).
򐂰 Some indexes were used for row selection (reason I1) or row selection and ordering or
grouping (reason I3).
򐂰 For some indexes, the data was retrieved from the index, without accessing the table
(QVC14 column).
Figure 4-26 Common columns from the 3001 record
84
4.6.5 The 3002 Record: Temporary index created
The 3002 record shows instances in which the database optimizer decided that existing
indexes are too costly or do not have the right key order for join, group by, or order by clauses.
For this reason, the optimizer might decide to create a temporary index (Classic Query
Engine (CQE)).
Index and table information
The following columns provide information about the index and table:
򐂰 QVQTBL and QVQLIB: Table name for which the index is built
򐂰 QQRCOD: Reason why the index build was done
I2
I3
I4
򐂰
򐂰
򐂰
򐂰
򐂰
Selection and ordering or grouping
Nested loop join
QQTOTR: Number of rows in the table
QQRIDX: Number of entries in the temporary index
QQSTIM: Time stamp for the start of the index build
QQETIM: Time stamp for the end of the index build
QQ1000: Name of the columns used for the index keys
Column names are the “short” column names.
record.
SELECT qqptln as "QQPTLN Library"
,qqptfn as "QQPTFN Table"
,qqrcod as "QQRCOD Reason"
,qqtotr as "QQTOTR Rows"
,qqridx as "QQRIDX Entries in Temp Idx"
,qqetim - qqstim as "Idx Build Time"
,qq1000 as "QQ1000 Index Created Key"
,qqidxa as "QQIDXA Index Advised"
,qqi2 as "QQI2 Nbr of Primary Key"
,qqidxd as "QQIDXD Index Advised"
,qvc16
FROM vldbmon
WHERE QQRID = 3002
ORDER BY "Idx Build Time" desc;
85
Figure 4-27 shows the result of the query in Example 4-9. It shows the following information:
򐂰 In the data sampled, we found some indexes that were created for a join were recognized
by reason code (QQRCOD) I4. Others were created for ordering or grouping
(reason code I2).
򐂰 No indexes were advised.
򐂰 Some rows show an index from an index. Notice the presence of information in columns
QQILNM and QQIFNM and the value on QVC16 (index from an index).
򐂰 Three rows show an index with a mapped key.
4.6.6 The 3003 Record: Query sort
The 3003 record shows that the database optimizer has decided to put selected rows into a
temporary space and sort them. This is either cheaper than alternative indexing methods or
an operation is forcing the optimizer to do so. An example is a UNION or an ORDER BY from
the columns of several tables.
The following columns are the most commonly used:
򐂰
򐂰
򐂰
򐂰
QQSTIM: Time stamp for the start of the refill and sort
QQETIM: Time stamp for the end of the refill and sort
QQRCOD and QQI7: Reason why a sort technique was chosen
QQRSS: The number of rows in a sort space
Keep in mind that sorting might increase the open time and cost since sorting is often
performed at open time. If a number rows sorted is small, then adding the right index might
improve performance. Indexes can still be used to select or join records before the sort
occurs. This does not indicate that the ODP is nonreusable.
The 1000 SQL summary record for the open might have a high elapsed time (QQI6 or QQI4).
Sort buffers are refilled and sorted at open time, even in reusable ODP mode. However, high
elapsed times might indicate a large answer set. In this case, the sort outperforms index
usage (the situation in most cases).
86
4.6.7 The 3004 Record: Temporary file
The 3004 record shows that the database optimizer is forced to store intermediate results and
rows in a temporary file because of the nature of the query. Examples are group by columns
from more than one file or materializing view results.
򐂰
򐂰
򐂰
򐂰
QQSTIM: Time stamp for the start of a fill temporary results table
QQETIM: Time stamp for the end of a fill temporary results table
QQTMPR: Number of rows in a temporary table
QQRCOD: Reason for building temporary index
record.
WITH xx AS
(SELECT * FROM qgpl.qzg0000478 WHERE qqrid = 3004),
yy AS
(SELECT qq1000 AS qqsttx, qqjfld, qqucnt FROM qgpl.qzg0000478 WHERE qqrid = 1000 AND qqc21
<> 'MT' AND (qvc1c = 'Y' OR (qqc21 IN('DL', 'UP')
SELECT qqstim as "Start timestamp" ,qqetim as "End timestamp",
DECIMAL((DAY(qqetim-qqstim)*24*3600)+
(HOUR(qqetim-qqstim)*3600)+(MINUTE(qqetim-qqstim)*60)+
(SECOND(qqetim-qqstim))+(MICROSECOND(qqetim-qqstim)*.000001),18,3) AS "Temp. population
time" ,qqrcod as "Reason" ,qqtmpr as "Rows in temp" ,qqsttx as "SQL text" FROM xx a LEFT
JOIN yy b ON a.qqjfld = b.qqjfld AND a.qqucnt = b.qqucnt ORDER BY "Temp. population time"
DESC;
Figure 4-28 shows the result of the query in Example 4-10. The information from the 3004
record is joined with information from the 1000 record so we know which queries were
implemented using a temporary result file. The reason in the example is that the query
contains a join condition that requires a temporary table (reason FA).
4.6.8 The 3006 Record: Access plan rebuild
The 3006 record is not present on every full open. It is generated only when an access plan
previously existed and has to be rebuilt. Also, the 3006 record is not generated when SQE
switches between cached access plans (up to three) in the SQE plan cache for an SQL
statement.
򐂰 QQRCOD: Rebuild reason code
򐂰 QQC21: Rebuild reason code subtype (IBM debug purposes)
򐂰 QVC22: Previous rebuild code
87
򐂰 QVC23: Previous rebuild code subtype
򐂰 QQC11: Plan required optimization
Y
N
Plan had to be re optimized & rebuilt
QAQQINI Re optimize option prevent access plan rebuild
The 1000 row contains an indicator regarding whether the rebuilt access plan can be
saved (QVC24).
record.
WITH xx AS
yy AS
(SELECT qq1000 AS qqsttx, qqstim, qqetim, qqjfld, qqucnt FROM qgpl.qzg0000478 WHERE qqrid
= 1000 AND qqc21 <> 'MT'
SELECT qqtime AS "Time", qqtim1 AS "Last Access Plan Rebuilt", qqc11 AS "Optimization
req.",
qqrcod AS "AP Rebuild Reason", HEX(qqc21) AS "Rebuild Subcode", HEX(qvc22) AS "Original
Rebuild Reason", HEX(qvc23) AS "Original Rebuild Subcode", varchar(qqsttx,20000) AS
"Statement Text"
FROM xx a LEFT JOIN yy b ON a.qqjfld = b.qqjfld AND a.qqucnt = b.qqucnt WHERE (a.qqtime
BETWEEN b.qqstim AND b.qqetim OR b.qqstim IS NULL) ORDER BY "Time";
Figure 4-29 show the result of the query in Example 4-11. The information from the 3006
record is joined with information from the 1000 record, so we know which queries had access
plans rebuilt. The example shows three reasons:
򐂰 The access plan was rebuilt because of system programming changes (reason A7).
򐂰 The storage pool changed, or the DEGREE parameter of CHGQRYA command changed
(reason AB).
򐂰 This is the first run of the query after a Prepare operation. That is, it is the first run with real
actual parameter marker values.
4.6.9 The 3007 Record: Index evaluation
The 3007 record shows all indexes that are evaluated for a given table, including which (if
any) were selected for use in this query and which were not and why. Reason codes are listed
next to each index. A reason code of 0 indicates that the index was selected.
This record also indicates whether the optimizer timed out while evaluating the indexes.
Indexes are evaluated in order from newest to oldest, in the same order as shown by the
DSPDBR CL command, excluding the views. To ensure that an index is evaluated, you can
delete and recreate it. This way it becomes first in the list.
88
򐂰
򐂰
򐂰
򐂰
QVQTBL: Table name
QVQLIB: Table library name
QQC11: Optimizer timed out (Y or N)
QQ1000: Contains library qualified index names, each with a reason code
– Reason code 0 indicates that an index was selected.
– Other codes are displayed in the second level text of CPI432C and CPI432D
messages.
– Documentation and iSeries Navigator reports classify this as the “Optimizer timed out”;
the 3007 row will be generated even when the optimizer does not time out.
record.
WITH xx AS
yy AS
SELECT qq1000 AS qqsttx, qqjfld, qqucnt
FROM qgpl.qzg0000478
WHERE qqrid = 1000 AND qqc21 <> 'MT'
SELECT qq1000,
varchar(qqsttx,20000) AS "Statement Text"
FROM xx a LEFT JOIN yy b
ON a.qqjfld = b.qqjfld AND a.qqucnt = b.qqucnt
Figure 4-30 shows the result of the query in Example 4-12. This is an example with
information from the qq1000 column in the 3007 record. The information is joined with
information from the 1000 record so we know the index evaluation for a given query. The first
row in the example shows that four indexes were evaluated and the last one (CUST_00003)
was used to implement the query.
Figure 4-30 Common columns from the 3007 record
4.6.10 The 3010 Record: Host variables
The 3010 record shows substitution values for host variables or parameter markers in the
query text (refer to the QQ1000 column in the 1000 record). This record appears just prior to
each instance of an open, update, delete, or insert with subselect. This record is not
displayed for insert with values. Data might not match exactly for updates with parameter
markers in the SET clause.
89
򐂰 QQ1000: This column contains substitution values for host variables or parameter marker.
The values (separated by commas) correspond left to right, with host variables and
parameter markers. All values are displayed as a character, with no special indication of
type. A floating point value is displayed as *F.
򐂰 QQUCNT and QQI5: These columns must be used to determine to which exact query the
substitution values belong. The 3010 row type is not generated for the INSERT with
Values statement.
4.6.11 The 3014 Record: General query optimization information
The 3014 record is displayed with full open optimization records. In most cases, one 3014
record is displayed per full open. You might see multiple 3014 records if the query consists of
multiple separately run queries, for example, a subquery with grouping functions or views that
need results materialized for use in the outer query. Values in this column help to identify the
type of query that this record represents and the amount of time it took to open the cursor for
this query.
This record also has summary information for the query or subquery. In most cases, there is
one 3014 row per full open. Subselects and materialized views can cause multiple 3014 rows.
It contains values for most of the settings that impact the query optimizer.
򐂰 QQC102: This column contains a library for the QAQQINI file if one is used. A value of *N
indicates that no QAQQINI file was used.
Many of the QAQQINI settings are found in the 3014 row; a couple are found in the 1000
row type.
򐂰 QVP154: Memory pool size
򐂰 QVP155: Memory pool ID
򐂰 QQC16: This column contains a Y when SQE is used to process the SQL statement
(CQE = N).
4.6.12 The 3015 Record: SQE statistics advised
The 3015 record is generated by SQE when it determines that a column statistic needs to be
collected or refreshed.
The following columns are commonly used:
򐂰 QVQTBL: Table name
򐂰 QVQLIB: Table library name
򐂰 QQC11: Statistics request type
N No statistic existed for the column
S Column statistic was stale
򐂰 QQ1000: Name of the column identified in Statistic Advice
Remember: Column statistics are created in the background automatically by default for
all Statistics Advised. Use QQUCNT to analyze the SQL request to determine if an index is
better than a column statistic.
4.6.13 The 3019 Record: Rows retrieved detail
The 3019 record shows a summary of the Fetch or Retrieve operations performed by DB2.
For non-SQL interfaces, such as OPNQRYF, it is the only way to determine number of rows
90
returned and the amount of time to retrieve those rows. This record can also be used to
analyze SQL requests.
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
QQI1: CPU time to return all rows, in milliseconds
QQI2: Clock time to return all rows, in milliseconds
QQI3: Number of synchronous database reads
QQI4: Number of synchronous database writes
QQI5: Number of asynchronous database reads
QQI6: Number of asynchronous database writes
QQI7: Number of rows returned
QQI8: Number of calls to retrieve rows returned
4.6.14 Record information for SQL statements involving joins
SQL statements with a JOIN are among the most sensitive areas in tuning SQL queries.
Unfortunately, there is not a specific record type for JOIN information. The information is
contained in the access method records used for the tables that are used in the query. To see
the join information, you need to look for the following, commonly used columns in records
3000 to 3004, 3007, 3021 to 3023, 3027 to 3029, among others:
򐂰 QQJNP: Join position
򐂰 QQC21: Join method, where NL indicates a nested loop, MF indicates a nested loop with
selection, and HJ indicates a hash join
Currently Database Monitor can only capture data related to a nested loop and nested
loop with selection.
򐂰 QQC22: Join type, where IN indicates an inner join, PO indicates a left partial outer join,
and EX indicates an exception join; also present in the 3014 record
򐂰 QQC23: Join operator
When it uses the nested loop with selection, it shows a Cartesian product, even when it is
not a Cartesian product.
򐂰 QVJFANO: Join fan out (Normal, Distinct fanout, or Unique fanout)
Example 4-13 shows a query to find information regarding the join operations.
Example 4-13 Information about join operations
SELECT qqrid AS "QQRID Record Type" ,qqc21 AS "QQC21 Join Method" ,qqjnp AS "QQJNP Join
Position" ,qqc22 AS "QQC22 Join Type" ,qqptln AS "QQPTLN Base Library" ,qqptfn AS "QQPTFN
Base Table" ,qqrcod AS "Reason"
,qqetim - qqstim AS "Elapsed Time" ,qqidxa AS "QQIDXA Index Advised" ,qqidxd AS "Suggested
keys"
FROM QZG0000479 a WHERE qqucnt = 28 AND qqjfld LIKE '%360528%' ORDER BY RRN(a)
Figure 4-31 shows the result of the query in Example 4-13. From the example, we can see:
򐂰 The operation joins two tables, CUST_DIM as the primary table and ITEM_FACT as the
secondary table.
򐂰 The type of join is an inner join (IN in the QQC22 column).
򐂰 The implementation for both tables was a table scan (QQRID is 3000).
򐂰 The reason for the table scan is because no indexes exist (T1 in the QQRCOD column).
򐂰 An index is advised on both tables (Y in the QQIDXA column).
򐂰 QQIDXD lists the suggested keys for the advised indexes.
91
Figure 4-31 Information about Join operations
4.6.15 New MQT record types
Enhancements have been added to the Database Monitor data to indicate usage of
materialized query tables (MQT).
Note: The query optimizer support for recognizing and using MQTs is available with V5R3
i5/OS PTF SI17164 and DB group PTF SF99503 level 4.
Although an MQT can contain almost any query, the optimizer only supports a limited set
of query functions when matching MQTs to user-specified queries. The user-specified
query and the MQT must both use the SQE. The optimizer only uses one MQT per query.
򐂰 1000/3006 Record: QQC22 has a new reason code of B5, which indicates that the access
plan needed to be rebuilt because the MQT was no longer eligible to be used. The
reasons might be that:
–
–
–
–
The MQT no longer exists
A new MQT was found
The enable or disable query optimization changed
Time since the last REFRESH TABLE exceeds the MATERIALIZED_QUERY_TABLE_
REFRESH_AGE QAQQINI option
– Other QAQQINI options no longer match
򐂰 3014 Record: This record shows the new QAQQINI file values:
– Logical field name QQMQTR, physical field name QQI7, contains the
MATERIALIZED_QUERY_TABLE_REFRESH_AGE duration. If the QAQQINI
parameter value is set to *ANY, the timestamp duration will be 99999999999999.
– Logical field name QQMQTU, physical field name QVC42. The first byte of QVC24
contains the MATERIALIZED_QUERY_TABLE_USAGE. Supported values are:
N *NONE, no materialized query tables are used in query optimization and
implementation.
A User-maintained refresh-deferred query tables might be used.
U Only user-maintained materialized query tables might be used.
򐂰 3000, 3001, 3002 Records: New columns have been added to the 3000, 3001, and 3002
records to indicate that a table or tables were replaced with an MQT. The logical field
name QQMQT, physical field name QQC13, is either Y or N, indicating that this is an MQT,
which replaced a table or tables in the query.
򐂰 3030 Record: The new 3030 record contains information about the MQTs that are
examined. This record is only written if MQTs are enabled and MQTs exist over the tables
specified in the query.
92
5
Chapter 5.
Analyzing database performance
data using iSeries Navigator
In Chapter 4, “Gathering database SQL performance data” on page 51, you learned about
the different Database Monitors and how to collect performance data. After you collect the
performance data, you must analyze it.
You can analyze the Database Monitor data by using predefined queries that come with
iSeries Navigator. In this chapter, we explain the analysis tool that is included with iSeries
Navigator and the predefined queries.
93
5.1 Considerations before analyzing Database Monitor data
You can collect Database Monitor data in several ways as explained in Chapter 4, “Gathering
database SQL performance data” on page 51. In some cases, you must import the data into a
graphical interface to take advantage of the reporting capabilities of the tool. This section
addresses some considerations for analyzing the Database Monitor data.
Note: Keep in mind that performance data can become quite large in size and could
consume a lot of disk space. The more data that you have, the more time it takes to
analyze it.
5.1.1 Importing the Database Monitor data
One way to collect Database Monitor data is to use the Start Database Monitor
(STRDBMON) CL command on a green screen. But to use the reporting capability of iSeries
Navigator, you must import the Database Monitor data. For ease of use, the structure and
layout of the performance data are identical to the data created by the SQL Performance
Monitor tool. This enables you to use the predefined reports in iSeries Navigator to analyze
performance data gathered by Database Monitor.
To analyze the data using the predefined reports, import the performance data collection from
the Database Monitor table into the SQL Performance Monitor. Select Database → SQL
Performance Monitors → Import as shown in Figure 5-1.
Figure 5-1 Selecting the options to import performance data from the Database Monitor file
94
In the Import SQL Performance Monitor Files window (Figure 5-2), you import your Database
Monitor performance data. Type the monitor name that you want to specify, select the schema
where you want the monitor to reside, and choose the type of monitor. Then click OK.
Figure 5-2 Import SQL Performance Monitor Files window
As shown in the example in Figure 5-3, the newly created SQL Performance Monitor now
reflects a status of Imported.
Figure 5-3 Imported SQL Performance Monitors
Note: It is possible to run the Database Monitor on one iSeries server and import the
collection data into a separate system to perform the analysis.
Chapter 5. Analyzing database performance data using iSeries Navigator
95
5.1.2 Reducing the analysis time
If you are collecting Performance Monitor information for all jobs or all queries, the Database
Monitors can generate a rather large file in the case of using a detailed SQL Performance
Monitor. Since the large monitor database files can cause slow response time during your
analysis, for faster analysis, you can select only those jobs in which you are interested by
using one of the following options:
򐂰 CRTDUPOBJ command and SQL INSERT with a subselect specifying the QQJNUM value
or values
򐂰 CREATE TABLE small as (SELECT...WHERE QQJNUM=xyz with DATA, where xyz is the
job number in which you are interested
Because the monitor files created from the Detailed SQL Performance Monitor have generic
names with sequence numbers as suffixes, you can find the monitor file name on the Saved
Data page of the Properties panel (Figure 5-4).
Figure 5-4 Saved Data properties of the Detailed SQL Performance Monitor
Note: After you copy the rows into a separate file, re-import the file again into the SQL
Performance Monitor.
You can also create indexes on the monitor file over common selection and grouping or order
by columns. For more information about index creation over commonly used keys, refer to
6.2.4, “Creating additional indexes over the Database Monitor table” on page 136.
96
5.2 Predefined database performance reports
Information gathered by the database monitoring tools provides the most complete pictures
about a query’s execution. You can analyze this information by using predefined result reports
in SQL Performance Monitor.
In general, there are two categories of performance data:
򐂰 Summary SQL Performance Monitoring: Summary SQL performance data is collected
via a Memory Resident Database Monitoring capability that can be managed with
application program interfaces (APIs) or with iSeries Navigator. When the monitor is
paused or ended, the data is written to disk so it can be analyzed. Because the
information is stored only in memory while it is collected, the performance impact on the
system is minimal, but the trade-off is that only summary data is gathered.
򐂰 Detailed SQL Performance Monitoring: Detailed SQL performance data is collected in
real time and recorded immediately to a database table. The monitor is not required to be
paused or ended to begin analysis of the results. Since this monitor saves the data in real
time, it might have a greater impact on overall system performance, depending on the
number of jobs that are monitored.
Note: The Memory Resident Monitor statistics are kept in main storage for fast recording,
but must be written to database files to use the iSeries Navigator interface to review the
results. To write the statistics to database files, either pause or end the started SQL
5.2.1 Accessing the collected performance data
To review the SQL Performance Monitor results, in the right pane, right-click the active SQL
Performance Monitor files. In the selection window that is displayed, you see a variety of
monitor actions, such as Pause, Continue, and End, as shown in Figure 5-5.
Figure 5-5 Managing the SQL Performance Monitor
97
From this list, you can choose from the following functions:
򐂰 Pause: This function stops the current collection of statistics and writes the current
statistics into several database files or tables that can be queried by selecting the Analyze
Results action. The monitor remains ready to collect more statistics, but requires the
Continue action to restart collection.
򐂰 Continue: This function restarts the collection of statistics for a monitor that is currently
paused.
Note: The Pause and Continue functions are activated only after a Summary type SQL
Performance Monitor is started because Detailed SQL Performance Monitor does not
need to be paused to view results.
򐂰 End: This function stops and ends the monitor. If you started a Summary type SQL
Performance Monitor, then it writes the current collection of statistics to the database files
or tables.
򐂰 Analyze Results: This function opens a window with three tabs for selecting ways to view
the collected statistics in the database files or tables:
– Memory Resident SQL Performance Monitor
•
•
•
Summary Results
Detailed Results
Composite View
– Detailed SQL Performance Monitor
•
•
•
Summary Results
Detailed Results
Extended Detailed Results
򐂰 List Explainable Statements: This function opens a window that lists the SQL
statements for which the Detailed SQL Performance Monitor has collected data and for
which a Visual Explain diagram can be produced. See 5.4.1, “List Explainable Statements”
on page 130, for an example.
򐂰 Properties: This function opens a window with three tabs that represent the original
monitor definition:
– General
– Monitored Jobs
– Saved Data
iSeries Navigator provides many predefined queries to view the recorded statistics. You can
select these queries by checking the various query types on the Analyze Results panels.
98
Memory Resident SQL Performance Monitoring
To begin viewing the results for the Memory-based Database Monitor, right-click the paused
or ended monitor and then select Analyze Results.
Figure 5-6 shows the first results panel that groups queries according to three tabs:
򐂰 Summary Results
򐂰 Detailed Results
򐂰 Composite View
Figure 5-6 Access SQL Performance Monitor results
On the Summary Results page, you can select individual queries or use the Select All button.
After you select the queries that you want to run, click the View Results button.
You can also choose to modify the predefined queries and run the new queries by clicking the
Modify Selected Queries button. For more information about this function, refer to 5.3,
“Modifying a predefined query report” on page 119.
The Memory Resident Database Monitor uses its own set of tables instead of using the single
table with logical files that the Detailed SQL Performance Monitor uses. The
Memory-Resident tables closely match the suggested logical files of the Detailed SQL
99
You can see the set of tables that correspond to the data to be collected by the Database
Monitor in the Saved Data property panel of Database Monitor as shown in Figure 5-7.
The monitor files have generic names with sequence numbers as suffixes. You can also see
the corresponding External Tables in Figure 5-7. For more information about the External
Table description for the Memory Resident Database Monitor, refer to Chapter 2, “Database
Monitor DDS” in DB2 Universal Database for iSeries Database Performance and Query
Optimization, which is available in the iSeries Information Center on the Web at:
Figure 5-7 Saved Data Property of Memory Resident Performance Monitor Results
Detailed SQL Performance Monitor
To access the collected Detailed Performance Monitor data, select File → Analyze Results
as shown in Figure 5-5. Notice that the Detailed SQL Performance Monitor is not required to
be paused or ended.
Figure 5-8 shows the first results window that groups queries according to three tabs:
򐂰 Summary Results
򐂰 Detailed Results
򐂰 Extended Detailed Results
100
Figure 5-8 Detailed SQL Performance Monitor results
You can select from a variety of predefined reports to view performance-related information
about your queries and choose reports that you want to analyze. Figure 5-9 shows an
example of a General Summary predefined report with summary information. For example, it
might indicate the number of times that tables are scanned and the number of times that the
optimizer recommends to create a permanent index.
Figure 5-9 General Summary report example
By analyzing the contents of these reports, you can identify problem areas in your queries
and take the appropriate steps to remedy such problems. The details of predefined reports
are described in the following section.
101
5.2.2 SQL performance report information from Summary reports
To begin viewing the summary reports from Memory Resident Database Monitor and the
Detailed SQL Performance Monitor results, click View Results in the Summary Results tab.
Because all reports from the summary results are summarized at a Database Monitor level,
you can get a high-level perspective of what was happening during the monitoring such as:
򐂰 Analysis of the type of SQL operation
򐂰 Identification of the jobs that consume the most DB2 resources
Summary reports of the Memory Resident Monitor
The Memory Resident Database Monitor collects much of the same information as the
Detailed SQL Performance Monitor, but the performance statistics are kept in memory. At the
expense of some detail, information is summarized for identical SQL statements to reduce the
amount of information collected. The objective is to get the statistics to memory as fast as
possible, while deferring any manipulation or conversion of the data until the performance
data is dumped to a result table for analysis.
Table 5-1 describes the summary reports that are available for the Memory Resident Monitor.
Table 5-1 Types of summary reports in the SQL Performance Monitor
Reports
Description
General
Summary
This report contains information that summarizes all SQL activity, such as the
amount of SQL used in the application, whether the SQL statements are mainly
short-running or long running, and whether the number of results returned is small
or large.
Job Summary
This report contains a row of information for each job. The information can be used
to tell which jobs on the system are the heaviest users of SQL, and therefore, which
ones are perhaps candidates for performance tuning.
Operation
Summary
This report contains a row of summary information for each type of SQL operation.
For example, if a large amount of INSERT activity is occurring, then using an
OVRDBF command to increase the blocking factor or perhaps use of the
QDBENCWT API is appropriate.
Program
Summary
This report contains a row of information for each program that performed SQL
operations. This information can be used to identify which programs use the most
or most expensive SQL statements. Note that a program name is available only if
the SQL statements are embedded inside a compiled program. SQL statements
that are issued through Open Database Connectivity (ODBC), Java Database
Connectivity (JDBC), or OLE DB have a blank program name unless they result
from a procedure, function, or trigger.
You can find the most time consuming job from a Job Summary report (see Figure 5-10). This
information is quite useful because a predefined Job Summary report can be a starting point
for further analysis.
Figure 5-10 Memory-Resident Job Summary report
102
Summary reports of the Detailed Performance Monitor
The Memory-Resident Monitor is not meant to replace the Detailed SQL Performance
Monitor. There are circumstances where the loss of detail in the Memory-Resident Monitor is
insufficient to fully analyze an SQL statement. In these cases, you should still use the
Detailed SQL Performance Monitor.
As shown for the Memory-Resident Monitor Summary report, in the summary report for the
Detailed Monitor, you can also identify the jobs that consume the most run time by viewing a
job summary report like the example in Figure 5-11.
Figure 5-11 Job Summary report in the Detailed Performance Monitor
From this report, you can see that the most time consuming job also has the highest value of
temporary index creation. You can also analyze the type of SQL operation from the Operation
Summary report as shown in Figure 5-12.
Figure 5-12 Operation Summary result in the Detailed Performance Monitor
In Figure 5-12, you can see that the most time-consuming SQL operation is OPEN. The
number of total table scans is 253, and the number of temporary index creation is 58. From
this report, you can see that the creation of perfect indexes can be one of many good
candidates for reducing OPEN time and the total run time.
103
Note: These reports are a good starting point, but they do not identify the cause of the
performance problem.
Table 5-2 describes additional reports of Detailed Performance Monitor results.
Table 5-2 Summary reports for the Detailed Performance Monitor
Reports
Description
SQL Attributes
Summary
This report contains a summary of the optimization attributes, which can help identify attributes
that potentially are more costly than others. For example, in some cases ALWCPYDTA(*YES) can
allow the optimizer to run a query faster if live data is not required by the application. Also,
*ENDMOD and *ENDPGM are much more expensive than *ENDJOB or *ENDACTGRP.
Isolation Level
Summary
This report contains a summary of the number of statements that were run under each isolation
level. The higher the isolation level is, the higher the chance of contention is between users. For
example, a high level of Repeatable Read or Read Stability use is likely to produce a high level of
contention. You should always use the lowest level isolation level that still satisfies the application
design requirement.
Error Summary
This report contains a summary of any SQL statement error messages or warnings that were
captured by the monitor.
Statement Use
Summary
This report contains a summary of the number of statements that are executed and the number
of times they are executed during the collection period of the Performance Monitor. This
information provides the user with a high-level indication of how often the same SQL statements
are used multiple times. If the same SQL statement is used more than one time, it might be
cached. Subsequent uses of the same statement are less expensive. It is more important to tune
an SQL statement that is executed many times than an SQL statement that is only run one time.
Open Summary
This report contains a summary of the number of statements that perform an open and the
number of times they are executed during the Performance Monitor collection period. The first
open of a query in a job is a full open. After this, the open data path (ODP) might be pseudo-closed
and then reused. An open of a pseudo-closed ODP is far less expensive than a full open. The
user can control when an ODP is pseudo-closed and the number of pseudo-closed ODPs are
allowed in a job by using the Change Query Attributes action in the Database Folder of iSeries
Navigator. In rare cases, an ODP is not reusable. High usage of nonreusable ODPs might indicate
that the SQL statements causing the nonreusable ODPs should be rewritten.
Data Access
Summary
This report contains a summary of the number of SQL statements that are read-only versus those
that modify data. This information provides the user with a less detailed view of the type of SQL
statements used than that available through the Operation Summary. This information can then
be used to try specific performance tuning techniques. For example, if a large amount of INSERT
activity is occurring, it might be appropriate to use the Override with Database File (OVRDBF)
command, to increase the blocking factor, or to use of the QDBENCWT API.
Statement Type
Summary
This report contains a summary of whether SQL statements are in extended dynamic packages,
in a system-wide statement cache, or in regular dynamic or static SQL statements. This
information provides the user with a high level indication of the number of SQL statements that
were fully parsed and optimized (dynamic). The information also indicates whether the SQL
statements and access plans were stored statically in a program, procedure, function, package,
or trigger. An SQL statement that must be fully parsed and optimized is more expensive than the
same statement that is static, extended dynamic, or cached in the system-wide statement cache.
Parallel Processing
Summary
This report contains a summary of the parallel processing techniques that were used. This
information provides the user with a high-level indication of whether one of the many parallel
processing techniques were used to execute the SQL statements. Most parallel processing
techniques are available only if the Symmetric Processing for iSeries is installed. When the option
is installed, the user must specify the degree of parallelism through the Change Query Attributes
action in the Database Folder of iSeries Navigator, the Change Query Attributes (CHGQRYA) CL
command, or the QQRYDEGREE system value.
104
Reports
Description
Optimizer
Summary
This report contains a summary of the optimizer techniques that were used. This information
provides the user with a high-level indication of the types of queries and optimizer attributes that
were used. You can use this information to determine whether the types of queries are complex
(use of subqueries or joins) and identify attributes that might deserve further investigation. For
example, an access plan rebuild occurs when the prior access plan is no longer valid or if a
change has occurred that identified a better access plan. If the number of access plan rebuilds is
high, it might indicate that some application redesign is necessary. Also, if the join order has been
forced, this might indicate that the access plan chosen is not the most efficient. However, it might
also indicate that someone has already tuned the SQL statement and explicitly forced the join
order because experimentation showed that a specific join order should always provide the best
order. Forcing the join order should be used sparingly. It prevents the optimizer from analyzing
any join order than the one specified.
For more information about SQL Summary Performance Monitor Summary results, refer to
the Preparing for and Tuning the V5R2 SQL Query Engine on DB2 Universal Database for
iSeries, SG24-6598.
5.2.3 SQL performance report information from Extended Detailed reports
The Extended Detailed results are most useful if the user has a basic knowledge of IBM
System i5™ or iSeries query optimization techniques. The most useful function from the
Extended Retailed reports is that you can analyze each SQL request.
Since each SQL statement can be identified by the monitor by according to job number,
statement number, and unique count, these reports give a DB2 trace at each SQL statement
level. But, because statements are not sorted or grouped by job, you must determine how to
address this issue as explained in the following sections.
Basic Statement Information report
From the Job Summary report shown in Figure 5-11 on page 103, where you can identify the
most time consuming job, we copied all data for a single job (job number 063229) into its own
collection to reduce analysis time. For more information about faster analysis, refer to 5.1.2,
“Reducing the analysis time” on page 96.
The most useful information is which query is the most time consuming. You can see that by
looking at the Basic Statement Information report (Figure 5-13).
Figure 5-13 The Basic Statement Information report
The Basic Statement Information report provides the user with basic information about each
SQL statement. The most expensive SQL statement is presented first in the list; at a glance,
the user can see which statements (if any) were long running.
Literals in the original SQL text might be replaced by parameter markers in prepared text if
SQL was able to convert them during preparation (desired). For the original SQL text, you can
105
use literal values from the matching Host Variable Values column in the place of parameter
markers.
The other important information from Basic Statement Information is Unique Count. In case
each query statement in the result report is sorted by only time stamp, you can identify a
specific query by seeking a Unique Count. The example in Figure 5-14 shows the Unique
Count of the most time consuming SQL statement.
Figure 5-14 The information of specific SQL statements
Since the Unique Count number does not increase any more if ODPs are reusable, then you
must also consider looking at the Refresh Count number to identify unique SQL statements
like the example in Figure 5-15.
Figure 5-15 Refresh Count with reusable ODPs
In our example, since ODP is reusable during repeated execution of same statement, Refresh
Count increases from 0 to 2, while Unique Count is fixed to 1011.
Optimizer Information report
From the previous information, you can see the Optimizer Information report (Figure 5-16).
This report provides you with basic optimizer information about SQL statements that involve
data manipulation (select, open, update and so on).
Figure 5-16 The Optimizer Information report
The Optimizer Information report contains a row of optimization information for each
subselect in an SQL statement. This information provides the user with basic optimizer
information about SQL statements that involve data manipulation (selects, opens, updates,
and so on). The most expensive SQL statement is presented first in the list.
In our example, you can see that the optimizer chose to use Classic Query Engine (CQE),
and ODP is reusable for this specific SQL statement. To retrieve a report of the specific SQL
106
statement more easily, you can also make the same report by clicking Modify Selected Query
and modifying the predefined query statement. In our example, a new selection clause is
inserted in the predefined query to retrieve a result of statement number 355 and Unique
Count 1015 as shown in Figure 5-17.
Figure 5-17 The report from Run SQL Script window
The same result is retrieved and displayed in the iSeries Navigator Run SQL Script window,
from where you can modify and execute it. For more information about modifying a predefined
query, see 5.3, “Modifying a predefined query report” on page 119.
Table Scan Information report
The Table Scan Information report indicates which entire table is scanned in arrival sequence
order without using an index. A table scan is generally acceptable in cases where a large
portion of the table is selected or the selected table contains a small number of records.
Otherwise, a table scan of large tables can be time consuming. Therefore, if you investigate a
long running SQL statement, using an index usually provides better performance.
To analyze and show table scan information, we run the sample query joining two tables as
shown in Example 5-1.
Example 5-1 Sample query joining two tables
select
from
where
and
a.orderkey, a. partkey, b.part, a.quantity
veteam21.item_fact a, veteam21.part_dim b
a.partkey = b.partkey
a.orderkey = '896';
107
We started SQL Performance Monitor in the Run SQL Script window using the JDBC
interface. The Table Scan Information report shows that the optimizer uses table scan for both
tables as shown in Figure 5-18.
Figure 5-18 Table Scan Information report
In the Table Scan Information report, the following information is the most commonly used:
򐂰 Reason Code: This column shows why the arrival sequence was chosen by the optimizer.
Our example in Figure 5-18 indicates No Index Exists in both of tables.
򐂰 Total rows in Base Table: This column indicates the number of rows in a table.
򐂰 Estimated Rows Selected: From this value, you can determine if the table scan was done
for a significant number of rows.
򐂰 Advised Index: This column indicates whether an index was advised (Yes or No).
򐂰 Advised Index Keys: This column suggests keys for an index. In our example, since both
tables do not have any indexes, the index creation can be one of good candidates for
providing better performance. Especially, since the table scan of ITEM_FACT table selects
only small portion of all rows, the index creation on ITEM_FACT is required more.
Note: The Advised Index column is set to “Y” only if the local selection key columns
exist on the table.
The Advised Index Keys column can contain both primary and secondary keys. The
Advised Primary Index keys column indicates the number of keys that are considered
primary. The other keys are considered less selective (secondary).
In case of ITEM_FACT table, two index key columns are advised. The first key column
(ORDERKEY) is for local selection, and the other (PARTKEY) is for a nested loop join with
the PART_DIM table. For the PART_DIM table, you can see that the Advised Index Key
column (PARTKEY) is for a nested loop join, although the Advised Index column is set to
“No”.
You can see the same results in Visual Explain diagram like the example shown in
Figure 5-19. This diagram is displayed by executing List Explainable Statements. For more
information about List Explainable Statements, refer to 5.4.1, “List Explainable Statements”
on page 130.
An insert with a subselect also has table scan information for the file inserted into, but this is
not a performance problem on its own. Nor does it indicate that the ODP is nonreusable. The
record data might contain useful Index Advisor data.
From this report, you can see that the perfect index creation is a key element for providing a
better access method with optimizer. Therefore, we investigate more information related to
the index.
108
Figure 5-19 Table Scan information from visual Explain
Open Information report
The Open Information report contains a row of information, for each open activity, for each
SQL statement. If an open operation occurs for the first time for a specific statement in a job,
it is a full open.
A full open creates an ODP that is used to fetch, update, delete, or insert rows. Since there
are typically many fetch, update, delete, or insert operations for an ODP, as much processing
of the SQL statement as possible is done during the ODP creation, so that the same
processing is not done on each subsequent I/O operation. An ODP might be cached at close
time so that, if the SQL statement is run again during the job, the ODP is reused. Such an
open is called a pseudo open, which is less expensive than a full open as shown in
Figure 5-20.
Note: At least, the same statement must be executed more than two times to create a
reusable ODP in the job.
Figure 5-20 Open Information report
Analyzing full opens is one of the most important steps because an ODP creation process is
expensive for query processing. The key point is determining why an SQL statement has an
excessive number of full opens.
In our example, since the sample query is executed only one time, you can see that the open
operation uses a full open that creates an ODP and the Full Opens column is set to “1”.
109
Otherwise, the Pseudo Opens column is set to “1”. The ODP Implementation column shows
that an ODP will be reusable if same SQL statement is executed more than two times.
If an ODP is not reusable, the Operation column is set to “CLOSE(Hard)” instead of “CLOSE”.
Since there are many reasons that the cursor is hard closed, you must look more carefully at
the Full Close Reason column.
Access Plan Rebuild Information report
The Access Plan Rebuild Information report contains a row of information for each SQL
statement that required the access plan to be rebuilt. Reoptimization is occasionally
necessary for one of several reasons such as a new index being created or dropped, applying
a program temporary fix (PTF), and so on. However, excessive access plan rebuilds might
indicate a problem.
In our example, you can see the reason why the access plan was rebuild as shown in
Figure 5-21.
Figure 5-21 Access Plan Rebuild Information report
From this report, the most important information is in the Access Plan Rebuild Reason
column. The following reasons are possible:
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
򐂰
110
Different File Or Member
Reusable Plan to Non-Reusable Plan Change
Non-Reusable To Reusable Plan Change
More Than Ten Percent Change In Number Of Rows
New Access Path Found
Access Path No Longer Found Or Valid
System Programming Change
Different CCSID
Different Date Or Time Format
Different Sort Sequence Table
Different Storage Pool or Paging Option
Symmetric Multi Processing Change
Different DEGREE
Open View or View Materialization
UDF or UDT Change
QAQQINI Change
Different Isolation Level or Scroll Option
Different FETCH FIRST n ROWS
First Run™ With Variable Values
Different Constraints
New Release
Index Advised Information report
Although it is possible to see the Advised Index keys from the Table Scan Information report,
you can see the same information from the Index Advised Information report as shown in
Figure 5-22.
Figure 5-22 Index Advised Information report
This report show that since there are no indexes on the ITEM_FACT table, the optimizer
chooses to use the Table Scan Data Access method. Notice that you can’t see index advised
information about PART_DIM table because the records included in this report are selected
only if the Advise Index column is set to “Y” in the Table Scan information report.
Visual Explain can offer more accurate information about an advised index. In our SQL
performance data, although you can’t see index advised information about PART_DIM table,
Visual Explain shows the Advised Index Key column information as shown in Figure 5-23.
Figure 5-23 Advised index key on table without selection key columns
For more information about Visual Explain, refer to Chapter 8, “Analyzing database
performance data with Visual Explain” on page 197.
Hash Table Information report
From the example in Figure 5-19 on page 109, you can see that the temporary hash table is
built on the ITEM_FACT table for hash prove. To analyze the cause of the hash table creation,
you can review the Hash Table Information report (see Figure 5-24).
This report shows information about any temporary hash tables that were used. The option is
available only when you use a detailed SQL Performance Monitor. Hash tables are used in
DB2 Universal Database for iSeries in two ways:
򐂰 For grouping
򐂰 For implementing the hash join algorithm
111
Figure 5-24 Hash Table Information report
Hash join and hash grouping might be chosen by the optimizer to perform an SQL statement
because it results in the best performance. However, hashing can use a significant amount of
temporary storage. If the hash tables are large, and several users are performing hash joins
or grouping at the same time, the total resources necessary for the hash tables might become
a problem.
Some useful columns are:
򐂰
򐂰
򐂰
򐂰
Hash table Entries
Hash Table Size
Hash table Row Size
Hash Table Pool Size
Figure 5-24 shows that the temporary hash table is created for the hash join.
Index Used Information report
According to the information from the previous reports, perfect indexes were created on both
tables. After creating new indexes, we ran the query again to see the effects of the indexes in
the query. From the newly created SQL Performance Monitor data, you can see that the
optimizer chooses to use existing indexes from the Index Used Information report as shown in
Figure 5-25.
Figure 5-25 Index Used Information report
This report shows the index that will be used to access the table and why it was chosen.
Since the index was chosen for a nested loop join, additional information is given to help
determine how the table fits in the join in Join Position and Join Method columns.
You can see same result in the Visual Explain diagram as shown in Figure 5-26. The index
probe access method was chosen by optimizer, although you cannot see the temporary hash
table any more in the Visual Explain diagram. The optimizer also does not advise an
additional index for better performance.
112
Figure 5-26 Visual Explain diagram showing effects of indexes
The following information is the most commonly used in this report:
򐂰 Reason Code: Why the optimizer chooses to use indexes, which might be for the following
possible reasons:
–
–
–
–
–
Selection only
Nested loop join
Record selection using bitmap
Bitmap selection
򐂰 Index or Constraint Name
򐂰 Index Type: Binary Radix Index or encoded-vector index (EVI)
򐂰 Join Position: Join position of each tables.
Index Create Information report
The Index Create Information report contains a row of information for each SQL statement
that required an index to be created. Temporary indexes might need to be created for several
reasons such as to perform a join, to support scrollable cursors, or to implement Order by or
Group by. The indexes that are created might contain keys only for rows that satisfy the query
(such indexes are known as sparse indexes). In many cases, index creation might be perfectly
normal and the most efficient way to perform the query. However, if the number of rows is
large, or if the same index is repeatedly created, you can create a permanent index to
improve performance of this query. This might be true regardless of whether an index was
advised.
A temporary index build does not always mean that the ODP is nonreusable. The database
optimizer tries to reuse the temporary index for each execution of the specific query if
possible.
For example, if host variable value is used to build selection into a temporary index (that is,
sparse), then ODP is not reusable because temporary index selection can be different on
113
every execution of the query. In this case, the optimizer tends to avoid creating sparse indexes
if the statement execution history shows it to be a frequently executed statement. But if the
temporary index build is done during repeated query run times and the query goes into
reusable ODP mode, then the temporary index is reusable.
Note: Temporary indexes are not usable by other ODPs.
If a particular query is run multiple times and a temporary index is built each time, a
permanent index must be created to avoid the index build and to make the ODP reusable.
Temporary indexes are never built for selection alone. They always involve a join or a group by
or order by clause. Furthermore, since no name is given to the temporary index, *TEMP is
used in subsequent monitor data.
To show an example, we use another sample query because the temporary index can be
created only if the optimizer chooses to use CQE for execution. Example 5-2 shows the
sample SQL statement.
Example 5-2 Sample query statement using CQE
select
from
where
and
and
and
group by
order by
a.orderkey, max(a.quantity) as qty
veteam23.item_fact a, veteam23.part_dim b
a.partkey = b.partkey
a.orderkey >= '800'
a.orderkey <= '900'
shipmode like 'T%'
a.orderkey
qty;
From the monitor data, you can see the Index Create Information report as shown in
Figure 5-27.
Figure 5-27 Index Create Information report
In this report, you can see that the Reason Code for index creation is the ORDER BY or
GROUP BY clause in SQL statement. But you can’t see any Advised Index information from
this report because a temporary index was built over a temporary table. If a temporary index
is built over a real table, you can see the advised index keys and create a perfect index using
advised keys.
You can also see the temporary index creation process from the Visual Explain diagram as
shown in Figure 5-28. In our example, the index build can’t be avoided if *MAP is one of the
keys listed, but this does not mean that the ODP is nonreusable. Notice that the purpose of
this example is only for getting the monitoring report. To tune this sample query, you must first
analyze the reason why temporary hash table was built on PART_DIM table.
114
Figure 5-28 Visual Explain diagram showing Temporary Index creation
Temporary File Information report
Figure 5-28 shows that temporary index is built over a temporary table. From the monitoring
data, you can see that the Temporary File Information report contains a row of information for
each SQL statement that required a temporary result.
Temporary results are sometimes necessary based on the SQL statement. If the result set
inserted into a temporary result is large, you might want to investigate why the temporary
result is necessary. In some cases, you can modify the SQL statement to eliminate the need
for the temporary result. For example, if a cursor has an attribute of INSENSITIVE, a
temporary result is created. Eliminating the keyword INSENSITIVE usually removes the need
for the temporary result, but your application sees changes as they occur in the database
tables.
In our example, you can see that the temporary file was built because a join condition
between two tables requires a temporary file for further processing as shown in Figure 5-29.
Figure 5-29 Temporary File Information report
115
Sort Information report
The Sort Information report contains a row of information for each sort that an SQL statement
performed. In our example, you can see that the optimizer chooses to use query sort
processing to obtain the final data as shown in Figure 5-30.
Figure 5-30 Query Sort report in the Visual Explain diagram
You can see the same results in the Sort Information report shown in Figure 5-31. Notice that
the reason for Sort is GROUP BY and ORDER BY Columns Different.
Figure 5-31 Sort Information report
Sorts of large result sets in SQL statement can be a time-consuming operation. In some
cases, an index can be created that eliminates the need for a sort.
Group By Information report
From the example in Figure 5-30, you can see that Index Grouping method is chosen by the
optimizer. You can see the information about the GROUP BY process in the Group By
Information Report as shown in Figure 5-32.
Figure 5-32 Group By Information report
This report shows that the temporary index is used for the GROUP BY process. This option is
available only when you use a detailed SQL Performance Monitor.
116
Other reports
The additional reports in the following sections are useful for query and SQL performance
analysis.
Governor Timeout Information report
The Governor Timeout Information report provides information about all optimizer timeouts.
This option is available only when you use a detailed SQL Performance Monitor. You can use
this report to determine how often users attempt to run queries that exceed the governor
timeout value. A large number of governor timeouts might indicate that the timeout value is
set too low.
Optimizer Timeout Information report
The Optimizer Timeout Information report provides information about any optimizer timeouts.
This option is available only when you use a detailed SQL Performance Monitor. Choosing the
best access plan for a complex query can be time consuming. As the optimizer evaluates
different possible access plans, a better estimate of how long a query takes shape.
At some point, for dynamic SQL statements, the optimizer might decide that further time
spent optimizing is no longer reasonable and use the best access plan up to that point. This
might not be the best plan available. If the SQL statement runs only a few times or if the
performance of the query is good, an optimizer timeout is not a concern. However, if the SQL
statement is long running, or if it runs many times and the optimizer times out, a better plan
might be possible by enabling extended dynamic package support or by using static SQL in a
procedure or program. Since many dynamic SQL statements can be cached in the
system-wide statement cache, optimizer timeouts are not common.
Procedure Call Information report
The Procedure Call Information report provides information about procedure call usage. This
option is available only when you use a detailed SQL Performance Monitor. Performance of
client/server or Web-based applications is best when the number of round trips between the
client and the server is minimized, because the total communications cost is minimized. A
common way to accomplish this is to call a procedure that performs a number of operations
on the server before it returns results, rather than sending each individual SQL statement to
the server. Figure 5-33 shows a sample report.
Figure 5-33 Procedure Call Information report
Distinct Processing Information report
The Distinct Processing Information report provides information about any DISTINCT
processing. This option is available only when you use a detailed SQL Performance Monitor.
SELECT DISTINCT in an SQL statement might be a time consuming operation because a
final sort might be necessary for the result set to eliminate duplicate rows. Use DISTINCT in
long running SQL statements only if it is necessary to eliminate the duplicate resulting rows.
Data Conversion Information report
The Data Conversion Information report contains a row of information for each SQL
statement that required data conversion. For example, if a result column has an attribute of
INTEGER, but the variable of the result is returned to DECIMAL, the data must be converted
from integer to decimal. A single data conversion operation is inexpensive. However, if the
117
operation is repeated thousands or millions of times, it can become expensive. In some
cases, it is easiest to change one attribute so a faster direct map can be performed. In other
cases, conversion is necessary because no exact matching data type is available.
In our example, you can see that data conversion is caused by Different Numeric Types in the
insert operation as shown in Figure 5-34.
Figure 5-34 Data Conversion Information report
Subquery Information report
The Subquery Information report contains a row of subquery information. This information
can indicate which subquery in a complex SQL statement is the most expensive.
Row Access Information report
The Row Access Information report contains information about the rows returned and I/Os
performed. This option is available only when you use a detailed SQL Performance Monitor.
This information can indicate the number of I/Os that occur for the SQL statement. A large
number of physical I/Os can indicate that a larger pool is necessary or that the Set Object
Access (SETOBJACC) command might be used to bring some of the data into main memory.
Lock Escalation Information report
The Lock Escalation Information report provides information about any lock escalation. This
option is available only when you use a detailed SQL Performance Monitor. In a few rare
cases, a lock must be escalated to the table level instead of the row level. This can cause
much more contention or lock wait timeouts between a user who is modifying the table and
the reader of the table. A large number of lock escalation entries might indicate a contention
performance problem.
Bitmap Information report
The Bitmap Information report provides information about any bitmap creations or merges.
This option is available only when you use a detailed SQL Performance Monitor. Bitmap
generation is typically used when performing index ANDing or ORing. This typically is an
efficient mechanism.
Union Merge Information report
The Union Merge Information report provides information about any union operations. This
option is available only when you use a detailed SQL Performance Monitor. UNION is a more
expensive operation than UNION ALL because duplicate rows must be eliminated. If the SQL
statement is long running, make sure it is necessary that the duplicate rows be eliminated.
118
Error Information report
The Error Information report provides information about any SQL statement error messages
and warnings that were captured by the monitor as shown in Figure 5-35.
Figure 5-35 Error Information report
If you receive an error message from your application, the error information report provides
SQLSTATE and SQLCODE. From these messages, you can analyze and correct errors.
Start and End Monitor Information report
The Start and End Monitor Information report provides information about any start and end
Database Monitor operations. This option is available only when you use a detailed SQL
5.3 Modifying a predefined query report
The SQL Performance Monitor also lets you retrieve the SQL statements of any of these
reports to use as a starting point for creating your own analysis reports. Although the
predefined queries can be materialized and customized, at first glance, they can be very
intimidating. Even power SQL users might be somewhat reluctant to make changes to these
statements.
This section explains how you can modify predefined queries and make your own reports.
Notice that the original query used by iSeries Navigator is not changed. You go step-by-step
through the process to customize a predefined query in iSeries Navigator.
119
1. From the Analyze Results window (Figure 5-36), select a query to modify. In this example,
we modify the Basic statement information query.
a. Click the Extended Detailed Results tab.
b. Under Select extended detailed queries, select the Basic statement information
query and click the Modify Selected Queries button.
Figure 5-36 Analyze Results window for customizing query
120
You see the SQL source for the query in the Run SQL Script window (Figure 5-37).
Figure 5-37 Basic statement information query on Run SQL Script window
121
2. In this example, we are trying to find top five most time-consuming queries from the report.
In the Run SQL Script window, scroll to the end of the query and add the following line
(see Figure 5-38):
FETCH FIRST 5 ROWS ONLY;
Click the Run Query icon from the toolbar (circled icon). The result appears at the lower
half of panel.
Figure 5-38 Modifying Pre-defined report
You can customize this report further. For example, you can omit columns from the report in
which you are not interested. You can also add grouping and totaling functions at the end of
the query. For example, if you want to know only the query statements and total runtime
accounting for the most runtime listed by each statements, you can customize query to insert
a grouping clause (by Total run time) and get a result like the one shown in Figure 5-39.
This query shows individual query instances and the amount of time they took, sorted by run
time (largest to smallest). It does not include non-ODP SQL operations such as prepare or
describe (QQUCNT=0 and QQSTN = 0). This provides a way to quickly find a query that is
taking a large amount of time with knowing any text about that query.
122
Figure 5-39 User Customized report
5.4 Query tuning example with SQL Performance Monitor
reports
Now we take you through a query tuning example using the Extended Detailed Reports.
There are two approaches to tuning the queries: a proactive approach and a reactive
approach. As the name implies, proactive tuning involves anticipating which queries will be
most often used for selection, joining, grouping, and ordering. In the reactive tuning, the
queries are tuned based on optimizer feedback, the query implementation plan, and system
performance measurements. In practice, both methods will be used iteratively.
As the numbers of users increase, more indexes are useful. Also, as the users become more
adept at using the application, they might start using additional columns that require more
indexes. Therefore, perfect index creation is the most effective tuning method.
The following example shows how the query is tuned by analyzing the Extended Detailed
Performance Monitor reports. This provides mainly a proactive approach for index creation.
For the purpose of this example, we assume that the sample query shown in Example 5-2 on
page 114 is the most time consuming and the object of query tuning. No index exists over
both tables used in sample query.
123
Figure 5-40 shows that the detailed SQL Performance Monitor is started and data is collected
from the Run SQL Script center.
Figure 5-40 Start SQL Performance Monitor
124
First we need to know the overall access plan chosen by the optimizer to find the starting
point for tuning. Since Visual Explain is the most useful tool to see an access plan, look at the
Visual Explain diagram in Figure 5-41, which shows a somewhat complex access plan.
Figure 5-41 Overall access plan for a sample query
First step
The first access method chosen by the optimizer is a table scan over both tables. To
understand why a table scan is chosen, you can see the Table Scan Information report
(Figure 5-42).
Figure 5-42 Table Scan Information report for tuning
This report shows that the reason for table scan is No Index Exists. According to the Advised
Index Keys column, we can choose index keys for the ITEM_FACT table. The index key
columns for local selection are SHIPMODE, ORDERKEY of the ITEM_FACT table.
125
Second step
Because the report from monitoring data can advise an index key only about Selection Key
columns, you can’t see the advised index keys for the PART_DIM table. Since the Join Key
columns can also be an Index Key column, we must consider information about the join
operation. From the Optimizer Information report (Figure 5-43), you can see that the hash join
method is chosen.
Figure 5-43 Optimizer Information report showing Hash join condition
The hash table from the PART_DIM table is created for the hash join operation as shown in
Figure 5-41. Since the Join Key columns of both tables are PARTKEY, you can select the Join
Key column of both tables as the Index Key column. Notice that hash table creation for the join
operation can be one of the useful candidates for index creation. The Index Key column for
join is the PARTKEY of the ITEM_FACT and PART_DIM tables.
From the results, we create indexes for both tables, and the order sequence of the columns is:
򐂰 ITEM_FACT: SHIPMODE, ORDERKEY, PARTKEY(IDX_TMP6)
򐂰 PART_DIM: PARTKEY(IDX_TMP7)
To see the result of applying new indexes, new Extended Detailed SQL Performance Monitor
data is collected again. After we apply new the indexes over both of the tables, we can’t see
records any more in the Table Scan Information report. However, the Index Used Information
report shows that the index scan method is chosen by optimizer as shown in Figure 5-44.
Figure 5-44 Index Used Information report after applying new indexes
The Optimizer Information report (Figure 5-45) also shows that the hash join is no longer used.
Figure 5-45 Optimizer Information report after applying indexes
126
You can see the same result from the Visual Explain diagram as shown in Figure 5-46.
Figure 5-46 Visual Explain diagram after applying indexes
Third step
Figure 5-46 shows that the optimizer still uses the Hash Grouping and Sort method.
Therefore, we must apply a new index for the grouping and sorting columns. You can see
information about each column from Group By information and Sort Information report.
From these reports, a new index is created over ITEM_FACT table and the key columns are
ITEM_FACT: ORDERKEY, QUANTITY(IDX_TMP8).
From the finally collected SQL Performance Monitor data, the Index Used Information report
shows that Hash Grouping is changed to Index Grouping as shown in Figure 5-47.
Figure 5-47 Index Used Information showing Index Grouping method
127
Although we applied the IDX_TMP8 index, the optimizer still uses the Sort method to get a
final result as shown in Figure 5-48.
Figure 5-48 Visual Explain diagram after applying the IDX_8 index
You can find the reason in the Sort Information report as shown in Figure 5-49.
Figure 5-49 Sort Information report after applying the IDX_8 index
From this report, we can see that the Sort method cannot be avoided because two different
columns are used for sorting.
128
Finally, we can see the OPEN process time from Basic Statement Information report as
shown in Figure 5-50.
Figure 5-50 Open process time after query tuning
Compared to the Basic Statement Information report, which shows the total run time before
applying indexes as shown in Figure 5-51, you can see that the OPEN process time has
decreased by less than 20 times.
Figure 5-51 OPEN process time before applying indexes
129
5.4.1 List Explainable Statements
The List Explainable Statements option from SQL Performance Monitor pane lists the SQL
statements for which a detailed SQL Performance Monitor has collected data and for which a
Visual Explain graph can be produced.
To access this function:
1. Click iSeries Navigator → Database → SQL Performance Monitors.
2. From the list of the SQL Performance Monitors that are currently on the system, right-click
a detailed collection and select List Explainable Statements, as shown in Figure 5-52.
Figure 5-52 Selecting the List Explainable Statements option
130
3. In the Explainable statements window (Figure 5-53), the upper half of the panel shows the
SQL statements that are monitored during the data collection session.
a. Click to select a statement that you are interested in analyzing. The selected statement
is then displayed in the lower half of the panel.
b. With the statement in focus, it is possible to have it analyzed and explained. Click the
Run Visual Explain button. For more information about “Visual Explain”, refer to
Chapter 8, “Analyzing database performance data with Visual Explain” on page 197.
c. With V5R3, since the columns in the upper half of the panel are sortable, you can
rearrange the data by clicking column head tab that you want to need to analyze. In our
example, we click the Time column to sort the columns by time sequence.
As you might have already noticed, the database analysis options and tools provided by
iSeries Navigator are well interconnected and meant to be used together in an iterative
manner.
Figure 5-53 Using List Explainable Statements
131
132
6
Chapter 6.
Querying the performance data
of the Database Monitor
In the previous chapters, we explained how the Detailed Database Monitor dumps all the
performance data into one table. We also explained the different columns in the table. Plus we
illustrated that the iSeries Navigator interface has some predefined reports that you can use
to understand and identify possible SQL performance problems.
In this chapter, we explain how to directly query the Database Monitor data. This is useful
because you can make your own queries that might not be available through the graphical
interface.
133
6.1 Introduction to query analysis
The Detailed Database Monitor table can be analyzed by using SQL. This is a time
consuming approach unless you have predefined queries. A lot of predefined queries exist via
the iSeries Navigator interface as explained in Chapter 5, “Analyzing database performance
data using iSeries Navigator” on page 93, but you can also write you own queries.
In this chapter, we present several queries to help you analyze the database performance
data. You can run the queries under all SQL interfaces that access the iSeries server. The
green-screen interface is intentionally not selected, because it can have a major negative
performance impact on some server models with reduced interactive capacity. All queries in
this chapter are run through the Run SQL Scripts window in iSeries Navigator.
To start the interface, in iSeries Navigator, select Databases → your database, right-click,
and select Run SQL Scripts (see Figure 6-1).
Figure 6-1 Starting the Run SQL Scripts option
134
6.2 Tips for analyzing the Database Monitor files
In the following sections, we present different tips to make it easier to analyze the
performance data using custom-made queries. The idea is to let you copy the different SQL
requests, so you can use them in your own analysis.
6.2.1 Using an SQL ALIAS for the Database Monitor table
By creating an SQL alias for the Detailed Database Monitor table that you are analyzing, you
can use the same names for the analysis. When you analyze the next Detailed Database
Monitor table, you use the SQL DROP ALIAS statement and then create an SQL ALIAS
statement with the same name over the other table.
If you want to use DBMONLIB.MYDBMON, then use the following SQL CREATE ALIAS
statement:
CREATE ALIAS DBMONLIB.MYDBMON FOR ibmfr.LAURA1608;
Before you analyze the next Database Monitor data, be sure to enter the SQL DROP ALIAS
statement:
DROP ALIAS DBMONLIB.MYDBMON
6.2.2 Using a subset of the Database Monitor table for faster analysis
The Database Monitor table often is large and contains information about many jobs.
Therefore, running queries on the data can sometimes be slower than desired. You can try to
reduce the time that the queries take by collecting only the job that you want. However,
sometimes this is not possible and, even if it is, batch jobs can generate a lot of Database
Monitor output. Also, using interactive tools, such as Start SQL (STRSQL), can result in
longer run times on server models.
If the response time is slow during the analysis, consider the following tips:
򐂰 Before you start the analysis, see how big the output table is for the collected Database
Monitor.
򐂰 Create a smaller table from the main Database Monitor table with only the rows in which
you are interested. You can use the following technique:
CREATE TABLE smaller-table AS (SELECT * FROM big-dbmon-table
WHERE QQJNUM IN(‘job-number’,’job-number’....) WITH DATA
򐂰 Another way to reduce the Database Monitor data is to include a time range in the SQL
selection criteria, for example:
and qqtime > '2005-03-16-12.00.00.000000' and qqtime < '2005-03-16-12.05.00.000000'
This shows only five minutes of collection.
You can adjust these techniques as needed to speed up your analysis.
Chapter 6. Querying the performance data of the Database Monitor
135
6.2.3 Using SQL views for the Database Monitor table
When using queries to analyze the Database Monitor table, you can make the queries more
readable by using views. For example, we look at a query that shows table scans, which can
be through a simple view, making it easier to see an overview of the query.
Example 6-1 shows the query before we create a view.
Example 6-1 Query before using an SQL view
WITH tablescans AS (
SELECT qqjfld,qqucnt,qqrest,qqtotr
FROM MYDBMON
WHERE qqrid=3000)
SELECT SUM(qqi6) "Total Time", COUNT(*) "Times Run",
a.qqucnt, integer(avg(b.qqrest)) "Est Rows Selected",
integer(avg(b.qqtotr)) "Total Rows in Table", qq1000
FROM MYDBMON a, tablescans b
WHERE qqrid=1000 AND a.qqjfld=b.qqjfld AND
qqc21 IN ('OP','SI','UP','IN','DL')
GROUP BY a.qqucnt, qq1000
ORDER BY "Total Time" DESC;
Then we create an SQL view as shown in Example 6-2.
Example 6-2 Creating an SQL view
CREATE VIEW LASSE0410.TABLESCANS AS SELECT QQJFLD, QQUCNT, QQRCOD
FROM MYDBMON WHERE qqrid=3000;
The query runs with the new view, with the assumption that the schema is in the library list as
shown in Example 6-3.
Example 6-3 Query after using the SQL view
In the remainder of this chapter, we do not use views, but rather table expressions. When you
analyze your own Database Monitor data collection, you might find situations where you can
benefit from using SQL views.
6.2.4 Creating additional indexes over the Database Monitor table
In Chapter 2, “DB2 Universal Database for iSeries performance basics” on page 9, we cover
the importance of indexes for SQL performance. The Database Monitor table tends to
become quite large in size. Therefore, it is important to create indexes on the Database
Monitor table over the common selection, grouping, and ordering clauses.
136
The following examples are some key combinations to use:
򐂰
򐂰
򐂰
򐂰
QQRID, QQC21
QQJFLD, QQUCNT, QQI5
QQRID & QQ1000
QQRID & QQC21
Figure 6-2 shows the SQL statements to create the indexes described previously.
Figure 6-2 Creating additional indexes for faster analysis
You can try other combinations as necessary. Remember to combine the selection, grouping,
and ordering clauses.
Note: When you use iSeries Navigator to start a detailed SQL Performance Monitor, as
soon as you end the data collection, it provides a couple of indexes, one based on
QQJFLD, QQUCNT & QQI5 and another based on QQRID & QQC21.
6.3 Database Monitor query examples
In this section, we present a series of queries to help solve specific questions in the detection
and resolution of SQL performance issues. Most of these queries have more elaborate
equivalents in the SQL Performance Monitor predefined reports as indicated. However, it is
still useful to be familiar with the Database Monitor table.
Before running the queries, we presume that an SQL SET SCHEMA and an SQL CREATE
ALIAS are done as shown in Figure 6-2.
137
6.3.1 Finding SQL requests that are causing problems
To find the SQL that is causing problems, expect to use more queries for the investigation,
because it is not only the running time that matters.
First take an overview over the data collected. To determine the number of SQL request that
were done during the collection, run the query shown in Example 6-4.
Example 6-4 Number of SQL requests in the Database Monitor table
SELECT count(*) FROM mydbmon
WHERE qqrid=1000 AND qqc21 <> 'MT';
Figure 6-3 shows the result of the query in Example 6-4.
Figure 6-3 Number of SQL requests in the DBMON collection
The result is smaller than the number of rows in the Database Monitor table. The number of
rows per SQL request can be from 1 to over 20, depending on the complexity of the SQL
request. Each SQL request has one QQRID = 1000 record. Sometimes the QQC21 row
contains the value MT for more text about the same SQL request, so it should only be
counted once.
For an overview of the most time consuming jobs running SQL, we use the query shown in
Example 6-5.
Example 6-5 Time consuming jobs
SELECT SUM(qqi6) "Total Time", COUNT(*) "Total SQL Requests",
qqjnum, qqjob, qquser FROM mydbmon
WHERE qqrid=1000 AND qqc21 <> 'MT'
GROUP BY qqjob,qquser,qqjnum ORDER BY 1 DESC;
Figure 6-4 shows the output of the query in Example 6-5.
Figure 6-4 Time consuming jobs
The total time is in microseconds. Therefore, you must divide the time by 1000000 to see it in
seconds.
138
To find SQL requests that might cause the problems, look at different information. For
example, you should know whether the Database Monitor collection is from a sample period
of the day or from a period when response problems were observed. You can also determine
which SQL requests in a specific job or jobs take the most processing time.
6.3.2 Total time spent in SQL
During the analysis of monitor data, you can see the percentage of time that is spent in DB2.
To begin, you find the start time and end time to have a duration of the Database Monitor data
collection as shown in Example 6-6.
Example 6-6 Duration of the Database Monitor data collection
SELECT MIN(qqtime) "Start Time", MAX(qqtime) "End Time",
MAX((DAY(qqtime)*24*3600)+(HOUR(qqtime)*3600)+(MINUTE(qqtime)*60)+
(SECOND(qqtime))+(MICROSECOND(qqtime)*.000001)) MIN((DAY(qqtime)*24*3600)+(HOUR(qqtime)*3600)+(MINUTE(qqtime)*60)+
(SECOND(qqtime))+(MICROSECOND(qqtime)*.000001)) AS "Duration"
FROM mydbmon WHERE qqrid<>3018
Figure 6-5 shows the result of Example 6-6.
Figure 6-5 Duration of Database Monitor data collection
To get the duration of the Database Monitor data collection for the job, select the job number
as shown in Example 6-7. For qqjnum = '999999', substitute the job number.
Example 6-7 Duration of the Database Monitor collection for one job
SELECT MIN(qqtime) "Start Time", MAX(qqtime) "End Time",
MAX(qqtime) - MIN(qqtime) "Duration (Sec)"
FROM mydbmon WHERE qqrid<>3018 and qqjnum = '999999';
You can also find the total number of seconds spent by using the SQL statement shown in
Example 6-8. The QQC21 has the value 'MT' when a More Text record exists.
Example 6-8 Time spent in SQL
SELECT SUM(qqi6)/1000000 "Total Time (Sec)" FROM mydbmon
WHERE qqrid=1000 AND qqc21 <> 'MT';
If stored procedures are used, then they count as double, because both the stored procedure
and the SQL requests in the stored procedure generate records in the Database Monitor
table. A good approximation is to exclude the stored procedure from the total time used in the
SQL shown Example 6-9.
Example 6-9 Time spent in SQL excluding stored procedures
SELECT SUM(qqi6)/1000000 "Total Time (Sec)" FROM mydbmon
WHERE qqrid=1000 AND qqc21 <> 'MT' AND qqc21 <> 'CA';
139
Figure 6-6 shows the output for Example 6-9.
Figure 6-6 Time spent in SQL
The time spent in SQL might not seem so relevant for the total run, but when the focus is on
individual jobs, or applications, then it is relevant. This means that a selection of the job
should be added to the query.
6.3.3 Individual SQL elapsed time
To find the SQL requests that count for the most run time, use the query shown in
Example 6-10.
Example 6-10 SQL request sorted on the total run time
SELECT SUM(qqi6) "Total Time" , COUNT(*) "Nbr Times Run", qq1000 "SQL Request"
FROM MYDBMON
WHERE qqrid=1000 AND qqucnt<>0
AND qqc21<>'MT'
GROUP BY qq1000 ORDER BY 1 DESC;
The result shown in Figure 6-7 comes from our test collection.
Figure 6-7 SQL requests sorted by Total Time
140
A large number of the same SQL request can give a high total run time. Therefore, it is also
relevant to look at the SQL requests with the longest average run time as shown in the query
in Example 6-11.
Example 6-11 SQL requests sorted by average run time
SELECT SUM(qqi6) "Total Time" , COUNT(*) "Nbr Times Run",
SUM(qqi6)/COUNT(*) "Average Run Time", qq1000 "SQL Request"
FROM MYDBMON
WHERE qqrid=1000 AND qqucnt<>0 AND qqc21<>'MT'
The execution of the query in Example 6-11 produces the result shown in Figure 6-8.
Figure 6-8 SQL requests sorted on Average Run Time
Important: Notice that the queries in Example 6-10 and Example 6-11 use GROUP BY in
the QQ1000 field whose length is 1000. If the monitor table contains SQL statements that
exceed this length and the first 1000 characters are identical, the result of the SUM and
COUNT functions will not be accurate.
6.3.4 Analyzing SQL operation types
To get an overview of the different SQL operation types that are run during the performance
data collection, use the query shown in Example 6-12.
Example 6-12 SQL operation types
SELECT SUM(qqi6) "Total Time", COUNT(*) "Nbr of Requests",
qqc21 "Operation Type" FROM MYDBMON
GROUP BY qqc21 ORDER BY 1 DESC;
141
Figure 6-9 SQL operation types
This example gives you an idea of the SQL operations that run the most and the amount of
time that they account for. In our example, there were 23768 OPEN operations.
6.3.5 Full open analysis
The first time (or times) that an open occurs for a specific statement in a job, a full open
operation is required. A full open creates an open data path (ODP) that is then used to fetch,
update, delete, or insert rows. An ODP might be cached at close time, so that if the SQL
statement is run again during the job, the ODP is reused. Such an open is called a pseudo
open and is much less expensive than a full open.
A normal SQL application has many fetches, inserts, updates, and deletes. A desirable
situation is that most of the operations share the ODP so that a full open does not have to be
done over and over again.
To find the number of SQL requests affected by full opens, you use the query shown in
Example 6-13.
Example 6-13 SQL requests affected by Full Opens
SELECT SUM(qqi6) "Total Time" , COUNT(*) "Nbr Full Opens", qq1000
FROM mydbmon
WHERE qqrid=1000 AND qqi5=0
AND qqc21 IN ('OP','SI', 'DL', 'IN', 'UP')
142
Figure 6-10 shows the results of the query in Example 6-13.
Figure 6-10 SQL requests affected by Full Opens
To analyze the full opens, you must next copy and paste the SQL request or part of the SQL
request into a query that we use for analysis. From the previous example, copy the following
SQL statement:
“SELECT T1.FIELD1, T1.FLAGS, T1.MINIMUMQUANTITY, T1.LASTUPDATE, T1.ENDDATE,
T1.TRADEPOSCN_ID, T1.QTYUNIT_ID, T1.STARTDATE, T1.OID, T1.MAXIMUMQUANTITY, T1.....”
Copy the SQL statement into the query shown in Example 6-14. Be sure to substitute the Xs
that follow the LIKE predicate.
Example 6-14 SQL requests affected by Full Opens in a single job
SELECT qqc181 "Cursor Name",qqi6 "Exec Time", qqucnt, qqi5, qqc21,
qqc15 "HC Reason", qq1000, qqc22 "Rebuild Reason Code",
qqc182 "Stmt Name"
FROM mydbmon
WHERE qqjnum='999999' AND
(qqrid = 1000 and qqc21 in ('OP') AND
qq1000 LIKE 'XXXXXXXXXXXXXXXXXXXXXXXXXXX%')
OR (qqc21 IN ('HC','DI', 'ST', 'CM', 'RO') )
ORDER BY qqtime;
143
Figure 6-11 shows the results for Example 6-14.
Figure 6-11 SQL requests affected by Full Opens in a single job
Look at the values in the QQC21 column for ODP analysis, which shows the following
abbreviations:
HC
DI
ST
CM
RO
Hard Close
Disconnect
Set Transaction
Commit
Rollback
Identify the reasons why a Hard Close is being done by looking in the QQC15 column, which
has these reason codes:
1
2
3
4
5
6
7
8
9
A
B
C
D
E
F
G
H
I
J
K
L
M
144
Internal Error
Exclusive Lock
Interactive SQL Reuse Restriction
Host Variable Reuse Restriction
Temporary Result Restriction
Cursor Restriction (after first execution)
Cursor Hard Close Requested (proprietary attribute)
Internal Error
Cursor Threshold
Refresh Error
Reuse Cursor Error
DRDA® AS Cursor Closed
DRDA AR Not WITH HOLD
Repeatable Read
Lock Conflict or QSQPRCED Threshold-Library
Lock Conflict or QSQPRCED Threshold-File
Execute Immediate Access Plan Space
Dummy Cursor Threshold
File Override Change
Program Invocation Change
File Open Options Change
Stmt Reuse Restriction
N
O
P
Internal Error
Library List Change
Exit Processing (End Connection)
Example 6-15 shows another query to find the number of Full Opens and Pseudo Opens for
the SQL request.
Example 6-15 Number of Full Opens and Pseudo Opens
SELECT SUM(qqi6) "Total Time" ,
SUM(CASE WHEN qvc12 = 'N' THEN 1 ELSE 0 END) "Full Opens",
SUM(CASE WHEN qvc12 = 'Y' THEN 1 ELSE 0 END) "Pseudo Opens",
QQ1000
FROM dbmon_file
GROUP BY qq1000
ORDER BY 1 DESC;
Figure 6-12 Number of Full Opens and Pseudo Opens
The total time in Figure 6-12 includes the Full Opens and Pseudo Opens for each SQL
request. To look at the time for Full Opens and Pseudo Opens separately, you must add
QVC12 to the GROUP BY clause from the previous query, as shown in Example 6-16.
Example 6-16 Number of Full Opens and Pseudo Opens looked separately
SELECT SUM(qqi6) "Total Time" ,
SUM(CASE WHEN qvc12 = 'N' THEN 1 ELSE 0 END) "Full Opens",
SUM(CASE WHEN qvc12 = 'Y' THEN 1 ELSE 0 END) "Pseudo Opens",
QQ1000
FROM dbmon_file
GROUP BY qvc12,qq1000
ORDER BY 1 DESC;
145
Figure 6-13 Full Opens and Pseudo Opens shown separately
To analyze the Full Opens, copy and paste the SQL request or part of the SQL request into a
query that used for analysis. From the previous query and result, we analyze the first SQL
request which is the most expensive. The first SQL request is similar to the third request.
Therefore, we use the LIKE predicate with two wildcards to ensure that we only retrieve
information for the first SQL request. Example 6-17 shows the query.
Example 6-17 Looking at the number of Full Opens and Pseudo Opens separately
SELECT qqc181 "Cursor Name",qqi6 "Exec Time", qqucnt, qqi5, qqc21,
qqc15 "HC Reason", qq1000, qqc22 "Rebuild Reason Code",
qqc182 "Stmt Name"
FROM dbmon_file
WHERE qqjnum='459263' AND
(qqrid = 1000 and qqc21 in ('OP') AND
UCASE(qq1000) LIKE 'SELECT SUM(QUANTITY)%SHIPMODE =%')
OR (qqc21 IN ('HC','DI', 'ST', 'CM', 'RO') )
ORDER BY qqtime;
Figure 6-14 Analysis of the Full Opens
From the previous example, we can see that a Full Open took place the first time. The
indication that it was a Full Open is the code in QQC21 and that QQI5 is 0. The Full Open
146
took 1.5 seconds. The query did not go into reusable mode and the cursor was hard closed
because of a cursor restriction (reason 6 in QQC15). The second Full Open took place again,
but this time a hard close didn’t occur leaving the ODP to be reused. Subsequent executions
reused the ODP. We can see this by looking at the QQUCNT and QQI5 fields. The number 8
in QQUCNT was assigned during the second Full Open and stayed constant for all
subsequent instances of that query. QQI5 has the number assigned to each instance of the
query. Notice that the execution time is minimum when the query entered into reusable mode.
For the complete list of statement types (QQC21) and the list of hard close reasons (QQC15),
search on Database Monitor: DDS in the V5R3 iSeries Information Center.
6.3.6 Reusable ODPs
Reusable ODP usually happens after the second execution of an SQL statement within the
connection or job, if the statement is reusable. Because the reusable statements are
significantly faster than the nonreusable ones, you can find the statements that are not
reusing the ODP. The QQUCNT value is assigned at full open time and stays constant for all
subsequent reusable instances of that particular query.
Nonreusable ODPs are indicated by the presence of optimization records each time a
particular query is run (full open). Reusable ODPs are indicated by 3010 and 1000 records
each time the given query is run (no optimization records or full open). To understand why an
ODP is not reusable, look at the hard close reason.
6.3.7 Isolation level used
You can see the number of statements that were run under each isolation level. This
information provides you with a high level indication of the isolation level used. The higher the
isolation level is, the higher the chance of contention is between users, which are seen as job
locks and seizes. A high level of Repeatable Read or Read Stability use is likely to produce a
high level of contention. Always use the lowest level isolation level that still satisfies the
application design requirement as indicated in the query shown in Example 6-18.
Example 6-18 Isolation level summary
select qvc41, count(qvc41) from mydbmon
WHERE QQRID = 1000 AND QQC21 <> 'MT'
GROUP BY qvc41 order by 2 desc;
Figure 6-15 Isolation level summary
147
The values can be translated using the following codes:
RR
Repeatable Read. In the SQL 1999 Core standard, Repeatable Read is called
serializable.
RS
Read Stability. In the SQL 1999 Core standard, Read Stability is called Repeatable
Read.
CS
Cursor Stability. In the SQL 1999 Core standard, Cursor Stability is called Read
Committed.
CSKL Cursor Stability KEEP LOCKS.
UR
Uncommitted Read. In the SQL 1999 Core standard, Uncommitted Read is called
Read Uncommitted.
NC
No Commit.
As you can see from the previous example, most reads are done using Cursor Stability. This
means that the isolation level is low, and therefore, the possibility for contention is low.
6.3.8 Table scan
A table scan operation is an efficient way to process all the rows in the table and verify that
they satisfy the selection criteria specified in the query. Its efficiency is accomplished by
bringing necessary data into main memory via a large I/O request and asynchronous
prefetches.
The table scan is generally acceptable in cases where a large portion of the table is selected
or the selected table contains a small number of records. To address cases where the entire
table is scanned, but a relatively small number of rows is selected, building an index on the
selection criteria is the best alternative and fully supported by the Database Monitor data.
Assuming that you have collected Detailed Database Monitor data, you can use the query
shown in the Example 6-19 to see the statements that have resulted in table scan operations.
Example 6-19 Table scan operations
SELECT qqjfld,qqucnt,qqrest,qqtotr
FROM MYDBMON
WHERE qqrid=3000)
148
Figure 6-16 Table scan operations
In the previous query, notice the following columns:
򐂰 QQJFLD and QQUCNT
These columns are join fields that are required to uniquely identify an SQL statement in
the Database Monitor file.
򐂰 QQC21
Since we are joining common table expression table scans back to the 1000 record in the
Database Monitor file, we must ensure that we only join to 1000 records that can cause
table scans to occur. We accomplish this by verifying that the QQC21 field operation is
either open, select, update, delete, or insert. We also include the last three operations
because they might have subselects or correlated subqueries.
򐂰 QQREST and QQTOTR
We included the QQREST (estimated rows selected) and QQTOTR (total rows in table)
columns to give you an idea of the selectivity of the statement. A great difference between
these two columns is a good indicator that index is a better alternative to a table scan
operation.
򐂰 QQRID
A table scan operation is uniquely identified by the record ID (QQRID) value of 3000. We
include it as selection criteria in the common table expression 'tablescans'.
򐂰 QQI6
This column indicates a table scan operation with a cumulative elapse time in
microseconds for each individual query. Since we use it as a cost indicator, we ordered the
output in descending order based on this value.
Note: Focus your efforts on optimizing statements that are displayed at the top. Ignore
statements whose total time was inflated due to numerous executions (times run field).
For other statements, consider the total number of rows in the table before taking
further action. If this number is relatively small, your optimization efforts are best spent
elsewhere.
149
The query that we have outlined so far is insufficient in helping us to decide if we should build
an index. Using the query shown in Example 6-19, we include data that is necessary to make
that decision as shown by Example 6-20.
Example 6-20 Keys advised
SELECT qqjfld,qqucnt,qqrest,qqtotr FROM MYDBMON WHERE qqrid = 3000),
details AS ( SELECT qqjfld, qqucnt, qqi7 FROM MYDBMON WHERE qqrid = 3019),
summation AS ( SELECT a.qqucnt, a.qqjfld, SUM(qqi6) "Total Time", COUNT(*) "Times Run",
integer(avg(b.qqrest)) "Est Rows Selected",
integer(avg(b.qqtotr)) "Total Rows in Table",
integer(avg(c.qqi7)) as "Rows Returned", qq1000
FROM MYDBMON a, tablescans b, details c
WHERE qqrid=1000 AND a.qqjfld=b.qqjfld AND a.qqjfld = c.qqjfld
AND qqc21 IN ('OP','SI','UP','IN','DL')
GROUP BY a.qqjfld, a.qqucnt, qq1000)
SELECT
a.qqtfn "File", a.qqtln "Library", a.qqi2 "Nbr of Primary Keys",
a.qqidxd "Keys Advised", a.qqrcod "Reason Code", b."Total Time",
b."Times Run", b."Est Rows Selected", b. "Rows Returned",
b."Total Rows in Table", b.qq1000
FROM
MYDBMON a, summation b WHERE a.qqjfld = b.qqjfld AND a.qqucnt = b.qqucnt AND
a.qqidxa = 'Y' AND a.qqrid = 3000 ORDER BY b."Total Time" DESC;
This query includes the following columns among others:
򐂰 QQI7
We included details for a common table expression to illustrate usage of the 3019 row type
- Detailed Row information. This row is written when *DETAILED is specified on Start
Database Monitor (STRDBMON) command or if using the Detailed option in iSeries
Navigator.
Note: Compare Rows Returned with the Total Rows in Table to gain an idea of the
selectivity for the query that caused the table scan operation.
You should also compare Estimated Rows Selected with Rows Returned. Consistently
great differences between these two columns are indicative that the query optimizer
needs better statistics (through additional statistics or indexes). The difference does not
indicate poor performance.
򐂰 QQTFN and QQTLN
These two columns refer to the base table and schema over which a table scan operation
was performed and for which we are considering building an index.
򐂰 QQIDXA
This column refers to the flag that specifies if an index was advised as indicated by Y or N.
We use it to filter out table scans for which no index was recommended by the query
optimizer (QQIDXA = 'Y').
򐂰 QQIDXD
This column indicates which index was advised. It lists the primary keys first, followed by
zero or more secondary keys. Secondary selection keys are less likely to have a
significant positive impact on the query’s performance.
򐂰 QQI2
This column indicates the number of primary keys contained in the QQIDXD column. You
can build an index over primary key fields for the most benefit.
150
򐂰 QQRCOD
This column indicates the reason code for a table scan operation. For individual reason
code descriptions, search on Database Monitor: DDS in the V5R3 iSeries Information
Center.
Note: Build indexes to reduce the cost of table scan operations where high selectivity
dictates usage of an index.
Always keep in mind that database performance optimization is an iterative process.
Therefore, after each action you take, recollect the Database Monitor data and re-analyze
it to validate that your action has resulted in better performance.
6.3.9 Temporary index analysis
A temporary index is a temporary object that allows the optimizer to create and use a radix
index for a specific query. The temporary index has all of the same attributes and benefits as
a radix index that is created by a user through the CREATE INDEX SQL statement or Create
Logical File (CRTLF) CL command. The temporary index can be used to satisfy a variety of
query requests, but it is only considered by the Classic Query Engine (CQE) when the query
contains ordering, grouping, or joins.
The created indexes might only contain keys for rows that satisfy the query (such indexes are
known as sparse indexes or select/omit logical files). In many cases, the index created might
be perfectly normal and the most efficient way to perform the query.
Look at the query in Example 6-21, which list temporary index builds ordered by the cost of
the index build time.
Example 6-21 Temporary index builds
SELECT
qqtotr
qqidxa
WHERE
qqucnt, qqetim-qqstim "Index Build Time", qvqtbl "Table Name", qvqlib "Schema",
"Rows in Table", qqridx "Entries in Index", SUBSTR(qq1000, 1, 100) "Key Fields",
"Index Advised", SUBSTR(qqidxd, 1,100) "Keys Advised" FROM MYDBMON
qqrid=3002 ORDER BY "Index Build Time" DESC;
151
Figure 6-17 Temporary index builds
We explain some of the columns used in the previous query:
򐂰 QQUCNT
This column uniquely identifies a query for a given job. It includes QQJFLD to uniquely
identify the query across many jobs.
򐂰 QVQTBL and QVQLIB
These columns indicate the long SQL table name and the long SQL schema name. Use
QQTFN and QQTLN for short object names.
򐂰 QQETIM-QQSTIM
The difference between ending and starting time stamps is represented as a decimal
(20.6). It is acceptable to use it as index build costing criteria, which is why our query uses
it in the ORDER BY clause.
򐂰 QQTOTR and QQRIDX
Compare the total rows in the table with the entries in the temporary index to gauge
whether the selection is built into a temporary index. If QQTOTR is greater than QQRIDX,
then the selection is built into a temporary index and you should carefully consider any
advised indexes located in QQIDXD field. In general, selection criteria keys should
precede QQ1000 keys, since QQ1000 fields are usually advised to satisfy join, ordering,
or grouping criteria.
򐂰 QQ1000
For this particular row type, the QQ1000 column contains key fields that are advised by
the query optimizer to satisfy criteria for join, ordering, grouping, scroll able cursors, and
so on. These are the reasons for the temporary index build. Therefore, use these keys in
any permanent index that you are going to build. Pay attention to any selection criteria
152
advised fields in QQIDXD and consider including them as the left-most key values in the
permanent index that you are going to build.
򐂰 QQIDXD
In the list of key columns in column QQIDXD, the optimizer lists what it considers the
suggested primary and secondary key columns. Primary key columns are columns that
should significantly reduce the number of keys selected based on the corresponding query
selection. Secondary key columns are columns that might or might not significantly reduce
the number of keys selected.
The optimizer can perform index scan-key positioning over any combination of the primary
key columns, plus one additional secondary key column. Therefore it is important that the
first secondary key column be the most selective secondary key column. The optimizer
uses index scan-key selection with any of the remaining secondary key columns. While
index scan-key selection is not as fast as index scan-key positioning, it can still reduce the
number of keys selected. Therefore, be sure to include the secondary key columns that
are fairly selective.
򐂰 QQRID
A row type of 3002 indicates a temporary index build so we have it in our selection criteria.
We modify the previous query to gather more information about the advised index and query
optimizer reasons for building the temporary index. Example 6-22 shows the modified query.
Example 6-22 Reason for building a temporary index
WITH qqq1000 AS (
SELECT qqjfld, qqucnt, qq1000 FROM
MYDBMON
WHERE qqrid = 1000 and qqc21 <> 'MT' AND
(qvc1c = 'Y' OR (qqc21 IN('DL', 'UP') AND qqc181 <= ' ') OR
qqc21 IN ('IN', 'IC', 'SK', 'SI') OR qqc21 LIKE 'O%'))
SELECT a.qqetim-a.qqstim "Index Build Time", a.qqrcod "Reason Code", a.qqtfn "File",
a.qqtln "Library", a.qqtotr "Rows in Table", a.qqridx "Entries in Index",
SUBSTR(a.qq1000, 1, 100) "Key Fields", a.qqidxa "Index Advised", a.qqi2 "Nbr of Primary
Keys", SUBSTR(a.qqidxd, 1,100) "Keys Advised", a.qvc16 "Index from index", b.qq1000
FROM MYDBMON a LEFT JOIN qqq1000 b ON a.qqjfld = b.qqjfld AND a.qqucnt = b.qqucnt
WHERE a.qqrid=3002 ORDER BY "Index Build Time" DESC;
153
Figure 6-18 Reason for building temporary index
We joined our initial query back to record 1000 of the base Database Monitor table to obtain
the SQL text for which the temporary index was built. The query uses the following columns:
򐂰 QQ1000
This column indicates that SQL statement that caused query optimizer to build a
temporary index.
򐂰 QQC21, QVC1C and QQC181
When joining back to row type 1000, we care only about the operations that can cause
query optimizer to advise an index. Therefore, we check for all the appropriate operation
types contained in field QQC21. Additional criteria might be contained in the QVC1C field,
SQL Statement Explainable, and QQC181, Cursor Name for the statement.
򐂰 QQRCOD
This column indicates the reason code for an index build. You most commonly see the
following reason codes:
I2
I4
I3
Ordering/grouping
Nested loop join
Selection and ordering/grouping
For a detailed list of possible reason codes, search on Database Monitor: DDS in the
V5R3 iSeries Information Center.
򐂰 QQI2
This column contains the number of suggested primary key columns that are listed in
column QQIDXD. These are the left-most suggested key columns. The remaining key
columns are considered secondary key columns and are listed in order of expected
selectivity based on the query. For example, assuming that QQIDXK contains a value of 4
154
and QQIDXD specifies seven key columns, then the first 4 key columns specified in
QQIDXK are the primary key columns. The remaining three key columns are the
suggested secondary key columns.
򐂰 QVC16
This column indicates that a temporary index was built from an existing index, which is
usually a short running operation.
Note: Building permanent indexes to replace temporary indexes can provide great returns
for a little time spent in analyzing Database Monitor data. Do not overuse this easy method
for short running and nonrepetitive temporary index builds.
6.3.10 Index advised
The query optimizer advises indexes and places the advise in the Database Monitor table
when the *DETAILED option is specified. The optimizer analyzes the row selection in the
query and determines, based on default values, if the creation of a permanent index improves
performance. If the optimizer determines that a permanent index might be beneficial, it
returns the key columns necessary to create the suggested index. Advised indexes can be
used to quickly tell if the optimizer recommends creating a specific permanent index to
improve performance.
While creating an index that is advised typically improves performance, this is not a
guarantee. After the index is created, much more accurate estimates of the actual costs are
available. Based on this new information, the optimizer might decide that the cost of using the
index is too high. Even if the optimizer does not use the index to implement the query, the new
estimates available from the new index provide more detailed information to the optimizer that
might produce better performance.
To look for the indexes advised by the optimizer, use the query shown in Example 6-23.
Example 6-23 Index advised
SELECT
qqucnt, qvqtbl "Table Name", qvqlib "Schema",
qqi2 "Nbr of Primary Keys",
SUBSTR(qqidxd, 1,100) "Keys Advised"
FROM MYDBMON
WHERE
qqrid IN (3000, 3001, 3002) and qqidxa='Y'
ORDER BY 5,2;
155
Figure 6-19 Index advised
This query uses the following columns:
򐂰 QQUCNT
This column uniquely identifies a query for a given job. Include QQJFLD to uniquely
identify a query across many jobs.
These columns indicate the long SQL table name and the long SQL schema name. Use
QQTFN and QQTLN for the short object names.
򐂰 QQIDXD
In the list of key columns contained in the QQIDXD column, the optimizer has listed what it
considers the suggested primary and secondary key columns. Primary key columns
should significantly reduce the number of keys selected based on the corresponding query
selection. Secondary key columns might or might not significantly reduce the number of
keys selected.
The optimizer can perform index scan-key positioning over any combination of the primary
key columns, plus one additional secondary key column. Therefore it is important that the
first secondary key column is the most selective secondary key column. The optimizer
uses index scan-key selection with any of the remaining secondary key columns. While
index scan-key selection is not as fast as index scan-key positioning, it can reduce the
number of keys selected. Therefore, include secondary key columns that are fairly
selective.
򐂰 QQI2
This column contains the number of suggested primary key columns that are listed in the
QQIDXD column. These are the left-most suggested key columns. The remaining key
columns are considered secondary key columns and are listed in order of expected
selectivity based on the query. For example, assuming that QQIDXK contains a value of 4
and QQIDXD specifies seven key columns, then the first 4 key columns specified in
QQIDXK are the primary key columns. The remaining three key columns are the
suggested secondary key columns.
156
򐂰 QQRID
Index advice is contained in three different row types and the query looks at all of them:
– 3000 - Table scan operation
We discussed the table scan operation previously as well as the need for indexes
where a query is highly selective.
– 3001 - Index used
In cases where an existing index was used, query optimizer might still recommend an
index. A recommended index might be better than the selected index, but not always.
Keep in mind that advised keys are for selection only and that you need to consider
JOIN, ORDER BY, or GROUP BY clause criteria.
– 3002 - Temporary index created
For temporary indexes, we recommend that you use a different query altogether
because we don’t illustrate the QQ1000 column for row type 3002 in this query. In this
case, QQ1000 includes keys used for a temporary index as well as their order
(ascending or descending).
򐂰 QQIDXA
We specified Y in this column since we are interested only in the rows for which query
optimizer has advised indexes.
Our initial query serves our need perfectly when we collect a very specific set of data (that is,
a single job). However, if you perform a system-wide Database Monitor collection, you must
use a query that is a bit more sophisticated.
We look at a query that enables us to perform costing by total run time required by the SQL
statements that caused indexes to be advised. Example 6-24 shows this query.
Example 6-24 Costing of SQL statements where an index is advised
WITH qqq1000 AS ( SELECT qqjfld, qqucnt, qq1000,
decimal(qqi6/1000000,21,6) AS "Total Runtime (sec)"
FROM MYDBMON
WHERE qqrid = 1000 AND qqi5 = 0 and qqc21 <> 'MT' AND (qvc1c = 'Y' OR (qqc21 IN('DL', 'UP')
AND qqc181 <= ' ') OR
qqc21 IN ('IN', 'IC', 'SK', 'SI') OR qqc21 LIKE 'O%'))
SELECT
b."Total Runtime (sec)", a.qqtfn "File Name",
a.qqtln "Library Name", a.qqi2 "Nbr of Primary Keys",
substr(a.qqidxd,1,100) "Keys Advised", b.qq1000
FROM
MYDBMON a LEFT JOIN qqq1000 b ON
a.qqjfld = b.qqjfld AND a.qqucnt = b.qqucnt WHERE
qqrid IN (3000,3001,3002) AND
qqidxa = 'Y' ORDER BY "Total Runtime (sec)" DESC;
157
Figure 6-20 Costing of SQL statements where an index is advised
We joined our initial query back to record 1000 of the base Database Monitor table to obtain
the SQL text and total statement run time. This query uses the following columns:
򐂰 QQ1000
This column indicates that an SQL statement caused query optimizer to advise an index.
򐂰 QQI6
This column indicates the total runtime required by the query, which is a good indicator of
the index benefit for a table scan and temporary index build operations. For the existing
index factors, such as join, grouping and ordering play into total runtime calculation, so this
field might not be an accurate cost indicator.
򐂰 QQC21, QVC1C and QQC181
When joining back to row type 1000, we are concerned about only the operations that can
cause query optimizer to advise an index. Therefore, we check for all the appropriate
operation types contained in column QQC21. Additional criteria might be contained in the
QVC1C column, SQL Statement Explainable, and QQC181, Cursor Name for the
statement.
The remainder of the statement is much like the initial query. One difference is that here we
order by total runtime of the SQL statement, which provides us with a good costing indicator.
This costing indicator helps us to focus on the worst performing statements and build indexes
intelligently.
158
You can see the number of different indexes that are advised and how many times they are
advised. To see this information, you run the query shown in Example 6-25.
Example 6-25 Listing of distinct index advised
SELECT distinct qvqtbl "Table", qqidxd "Key Fields", count(*) "Times adviced"
FROM MYDBMON
WHERE qqrid IN (3000, 3001, 3002) and qqidxa='Y'
group by qvqtbl, qqidxd
order by qvqtbl, qqidxd;
Figure 6-21 Different index advised
This is a good example where creation of the index with the key field PUBLISHED is not
necessary, because creation of the index with the key fields PUBLISHED, LANGU00001
covers both index recommendations, with a total of 33 times advised.
6.3.11 Access plan rebuilt
An access plan consists of one or more integrated steps (nodes) that are assembled to
retrieve and massage data from DB2 tables to produce results that are desired by the
information requestor. These steps might involve selecting, ordering, summarizing, and
aggregating data elements from a single table or from related (joined) rows from multiple
tables.
Each SQL query executes an access plan to retrieve the data that you requested. If the
access plan does not exist already, the system builds one dynamically, adding overhead to
the total time required to satisfy your request.
As a general rule, we want to avoid access plan rebuilds. That said, there are several perfectly
valid reasons for access plan rebuilds, for example:
򐂰 Deleting or recreating the table to which the access plan refers
򐂰 Deleting an index that is used by the access plan
򐂰 Applying database PTFs
򐂰 Table size changing by 10%
򐂰 Creating a new statistic, automatically or manually
򐂰 Refreshing a statistic, automatically or manually
򐂰 Removing a stale statistic
򐂰 CQE rebuilding an access plan if there was a two-fold change in memory pool size
159
򐂰 SQL Query Engine (SQE) looking for a ten-fold change in memory pool size if memory
pool is defined with a pool paging option of *CALC (also known as expert cache) (If paging
is set to *FIXED, SQE behaves the same as CQE.)
򐂰 Specifying REOPTIMIZE_ACCESS_PLAN (*YES) in the QAQQINI table or in the SQL
script
򐂰 Specifying REOPTIMIZE_ACCESS_PLAN (*FORCE) in the QAQQINI table or in the SQL
script
򐂰 Changing the number of CPUs (whole or fractions using logical partition (LPAR)) that are
available to a query results in rebuilding the access plan.
򐂰 Access plans are marked as invalid after an OS/400 release upgrade.
򐂰 The SQE Plan Cache is cleared when a system initial program load (IPL) is performed.
The SQE access plan rebuild activity takes place below the machine interface (MI).
Therefore, compared to CQE, you should see much less performance degradation caused by
lock contention on SQL packages, caches, and program objects.
At times, even though the optimizer rebuilt the access plan, the system fails to update the
program object. The most common reason for this failure is that other jobs are using the same
program and optimizer cannot obtain the exclusive lock on the program object to save the
rebuilt access plan. Another reason is that the job does not have proper authority to the
program or the program is currently being saved. The query still runs, but access plan
rebuilds continue to occur until the program is updated.
The rebuilt access plan might be saved in the existing access plan space within the program,
SQL package, or cache. If a new access plan is greater than the existing one, new space is
allocated and the plan is saved in that newly allocated space. If the number of access plan
rebuilds is high, some application redesign might be necessary.
We look at access plan rebuild information that is available in Database Monitor data by using
the query shown in Example 6-26.
Example 6-26 Access plan rebuilds information
WITH rebuilds AS (
SELECT qqjfld, qqucnt, qqrcod
FROM
MYDBMON
WHERE qqrid=3006 )
SELECT
a.qqucnt, b.qqrcod "Rebuild Reason",
qvc24 "Plan Saved Status", qq1000
FROM
MYDBMON a, rebuilds b
WHERE
a.qqjfld=b.qqjfld AND a.qqrid=1000 AND
a.qqc21 NOT IN ('MT','FE','CL','HC')
ORDER BY 4, 1;
160
Figure 6-22 Access plan rebuilds information
򐂰 QQRID
Our common table expression rebuild contains only row type 3006, which has information
specific to access plan rebuilds. Row type 3006 is not present on every full open. It is only
generated when an access plan previously existed and now has to be rebuilt.
Row type 3006 is also not generated when SQE switches between cached access plans
(up to three) in the SQE plan cache for an SQL statement.
By now you know that QQJFLD and QQUCNT are join fields required to uniquely identify
the SQL statement in the Database Monitor file.
򐂰 QQRCOD
This column provides the reason code for the access plan rebuild. There are over twenty
possible reason codes. For a detailed description of the reason codes, search on Database
Monitor: DDS in the V5R3 iSeries Information Center.
򐂰 QVC24
This column indicates the reason code for why the access plan was saved or not saved.
All A* codes mean that the access plan was not saved. AB means that a lock could not be
obtained, A6-A9 means that not enough space was available, and AA means that a plan
was saved by another job.
All B* codes mean that the access plan was saved, with a blank value or B3, B4, or B6
meaning that a plan was saved “in place”, and B1, B2, B5, B7, or B8 meaning that the plan
was saved in a “new” space.
161
For a detailed description of each reason code, search on Database Monitor: DDS in the
򐂰 QQ1000
This column indicates the SQL statement that caused query optimizer to rebuild the
access plan.
򐂰 QQC21
In this query field, QQC21 is pulled from row type 1000 and represents the operation type
for the SQL statement. We exclude continuation records, fetches, closes, and hard closes.
The output is in ascending order based on the statement text and the unique statement
counter. This query provides basic information about access plan rebuilds.
Let us rewrite the query to include more information related to access plan rebuilds as shown
in Example 6-27.
Example 6-27 Extended access plan rebuild information
WITH rebuilds AS (
SELECT qqjfld, qqucnt, qqrcod, qqc21, qqc11, qqtim1
FROM MYDBMON
WHERE qqrid=3006 ) SELECT
b.qqrcod "Rebuild Reason",
hex(b.qqc21) "Reason Subcode (for IBM debug)", a.qvc24 "Plan Saved Status",
a.QQC103 "Package/Program Name",
a.QQC104 "Package/Program Library", a.qvc18 "Statement Type", b.qqc11 "Plan Reoptimized",
b.qqtim1 "Last Rebuilt", a.qq1000
FROM MYDBMON a, rebuilds b WHERE a.qqjfld=b.qqjfld AND a.qqucnt = b.qqucnt and a.qqrid=1000
AND a.qqc21 NOT IN ('MT','FE','CL','HC') ORDER BY "Rebuild Reason", "Plan Saved Status";
Figure 6-23 Extended access plan rebuild information
162
The query uses the following columns:
򐂰 QQC21
In this query, we pull the QQC21 column from row type 3006. In this row type, field QQC21
contains the access plan rebuild reason subcode. This subcode is in hex and should only
be used when reporting a problem to IBM Support.
We still use QQC21 from row type 1000 to exclude undesired 1000 records, like we did in
the initial query.
򐂰 QQC11
This column contains a simple character flag that indicates if an access plan required
optimization. If the value is Y, then the plan was reoptimized. If the value is N, then the
plan was not reoptimized. If this value is never Y, it is possible that QAQQINI file contains a
REOPTIMIZE_ACCESS_PLAN setting that prevents the query optimizer from
reoptimizing the access plan.
򐂰 QQTIM1
The value in this column indicates the time stamp of last access plan rebuild.
򐂰 QQC103
This column contains the name of the package or program that contains the SQL
statement that caused query optimizer to rebuild the access plan.
򐂰 QQC104
This column indicates the name of the library that contains the program or package listed
in the QQC103 column.
򐂰 QVC18
This column describes query statement type. If statement was done as part of dynamic
processing, it is flagged with E for “extended dynamic”, with S for “system wide cache”,
and L for “local prepared statement”.
6.3.12 Query sorting
The 3003 record from the Database Monitor table shows that the database optimizer has
decided to place selected rows into a temporary space and sort them. The presence of a
3003 record does not necessarily indicate poor performance. The optimizer selected a query
sort because it is either cheaper than the alternative indexed methods or it is forced to do so,
for example when UNION is used or ORDER BY uses columns from more than one table.
Indexes can still be used to select or join rows before the sort occurs. The 3006 record does
not indicate that the ODP is nonreusable.
Sort buffers are refilled and sorted at open time, even in reusable ODP mode.
Sorting might increase the open (OP) time/cost since sorting is often performed at open (OP)
time. This means that it might take some time to return the first row in the result set to the end
user.
High elapsed times for a query sort might indicate a large answer set. In this case, the sort
outperforms index usage (the situation in most cases). You should not attempt to build
indexes for queries with large result sets, unless you are going to add selection criteria to the
SQL statement’s WHERE clause to further reduce the result set.
If the answer set is small, but the optimizer does not have the right indexes available to know
that, creating indexes over selection columns can help by giving the optimizer statistics and
163
an alternative method of accessing the data. This is possible only if the optimizer is not forced
to use the sort (that is via a UNION or ORDER BY on columns from more than one table).
Look at which queries involve the use of a query sort. You can do a query sort by using the
query shown in Example 6-28.
Example 6-28 Use of a query sort
WITH sorts AS (
SELECT qqjfld, qqucnt
FROM mydbmon
WHERE qqrid=3003 )
SELECT SUM(qqi6) "Total Time" , COUNT(*) "Nbr Times Run", a.qqucnt, qq1000
FROM mydbmon a, sorts b
WHERE qqrid=1000 AND a.qqjfld=b.qqjfld and a.qqucnt=b.qqucnt AND
Figure 6-24 Use of a query sort
򐂰 QQRID
Our common table expression sort contains only row type 3003, which has information
specific to SQL statements using query sorts.
The QQJFLD and QQUCNT are join fields required to uniquely identify the SQL statement
in the Database Monitor file.
򐂰 QQC21
Since we are joining common table expression sorts back to the 1000 record in the
Database Monitor file, we must ensure that we only join to 1000 records that can cause
164
query sorts to occur. This is accomplished by verifying that QQC21 field operation is either
open, select, update, delete, or insert. We need to include last three operations because
they might have subselects or correlated subqueries.
򐂰 QQI6
This column indicates a table scan operation cumulative elapse time, in microseconds for
each individual query. Since we use it as a cost indicator, we have ordered the output in
descending order based on this value.
The query that we have outlined so far is insufficient in helping us to decide if building an
index or modifying the SQL statement is desired. Therefore, we revise the query (see
Example 6-29) to include data that is necessary to make the decision if any action is possible.
Example 6-29 Including data showing possible action
WITH sorts AS (SELECT qqjfld, qqucnt FROM mydbmon WHERE qqrid=3003 ),
summation AS (SELECT SUM(qqi6) "Total Time" , COUNT(*) "Nbr Times Run", a.qqjfld,
a.qqucnt, qq1000 FROM mydbmon a, sorts b WHERE qqrid=1000 AND a.qqjfld=b.qqjfld AND
a.qqucnt = b.qqucnt and qqc21 IN ('OP','SI','UP','IN','DL') GROUP BY a.qqjfld, a.qqucnt,
qq1000), fetches AS (SELECT a.qqjfld, a.qqucnt, integer(avg(a.qqi3)) "Rows Fetched" FROM
mydbmon a, summation b WHERE qqrid=1000 AND a.qqjfld=b.qqjfld and a.qqucnt = b.qqucnt AND
qqc21 = 'FE' GROUP BY a.qqjfld, a.qqucnt) SELECT b."Total Time", b."Nbr
Times Run",
a.qqrcod "Reason Code", a.qqi7 "Reason subcode for Union",
a.qqrss "Number of rows sorted", c."Rows Fetched",
a.qqi1 "Size of Sort Space", a.qqi2 "Pool Size",
a.qqi3 "Pool ID", a.qvbndy "I/O or CPU bound", a.qqucnt, b.qq1000 FROM
summation b LEFT OUTER JOIN fetches c ON b.qqjfld = c.qqjfld AND
b.qqucnt = c.qqucnt INNER JOIN mydbmon a
ON b.qqjfld = a.qqjfld AND b.qqucnt = a.qqucnt WHERE
a.qqrid = 3003 ORDER BY
b."Total Time" DESC;
Figure 6-25 shows the output from the query in Example 6-29.
Figure 6-25 Include data showing action possible
165
򐂰 QQRCOD
This column indicates the reason for choosing the query sort technique. The value in this
column helps to identify whether the sort required of the query optimizer determined that
the cost of the sort is better than any other implementation (such as an index).
If you can change the SQL statement itself, any reason code is available for optimization
efforts. Or perhaps you cannot change the SQL statement (that is to optimize the
third-party ERP application) and can only build indexes and change other environmental
factors to help performance (that is, increase the pool size). In this case, focus your
optimization efforts on query sorts with reason code F7 (optimizer chose sort rather than
index due to performance reasons) and F8 (optimizer chose sort to minimize I/O wait
time).
For a detailed description of each reason code, search on Database Monitor: DDS in the
򐂰 QQI7
This column indicates the reason subcode for the UNION clause. If the query sort reason
code lists F5 (UNION was specified for query), this column contains one of two subcodes.
A value of 51 means that there is also an ORDER BY in the statement. A value of 52
means that the query specifies UNION ALL rather than simply UNION.
򐂰 QQRSS
This column tells us the number of rows that are contained in the sort space. You can use
this value, along with the reason code, to determine if the indexed approach is possible
and possibly cheaper (for a small result set). Compare the value of QQRSS with the value
of QQI3 from the corresponding 1000 FE record for this open to determine the number of
rows that were fetched from the sort space.
If the number of rows in sort space is large, but the actual number of rows fetched is small,
consider adding OPTIMIZE FOR n ROWS to the query to help the optimizer make a better
decision.
Building a more perfect index for the selection criteria might also help the optimizer make a
better decision and use index for the implementation method rather than a query sort.
򐂰 QQI3 from row type 1000 operation id FE (fetch)
This column tells us the number of rows that were fetched from the sort space to satisfy a
user request. As described in the QQRSS column description, the value in this column is
used to gauge whether more information is required by the query optimizer to make better
costing decisions.
򐂰 QQI1
This column indicates the size of the sort space.
򐂰 QQI2
This column indicates the pool size.
򐂰 QQI3 from row type 3003
This column indicates the pool ID.
򐂰 QVBNDY
This column contains a flag that indicates whether the query sort is CPU or I/O bound.
We have taken the base query and modified it to include more information about the query
sort implementation. This additional information helps you make more intelligent decisions
166
when deciding to optimize SQL statements using query sorts as the implementation method.
The most valuable new columns indicate a reason code and the number of actual rows
fetched for the query.
Changing your SQL statement or adding the OPTIMIZE FOR x ROWS syntax is most likely to
help alleviate issues that pertain to long query sort times. For highly selective queries where
sort space is disproportionately larger than actual rows fetched, building a more perfect index
might help the optimizer.
6.3.13 SQE advised statistics analysis
With the introduction of SQE to OS/400 V5R2, the collection of statistics was removed from
the optimizer and is now handled by a separate component called the Statistics Manager. The
Statistics Manager has two major functions:
򐂰 Create and maintain column statistics
򐂰 Answer questions that the optimizer asks when finding the best way to implement a given
query
These answers can be derived from table header information, existing indexes, or
single-column statistics. Single-column statistics provide estimates of column cardinality,
most frequent values, and value ranges.
These values might have been previously available through an index, but statistics have the
advantage of being precalculated and are stored with the table for faster access. Column
statistics stored with a table do not dramatically increase the size of the table object. Statistics
per column average only 8 to 12 KB in size. If none of these sources are available to provide
a statistical answer, then the Statistics Manager bases the answer on default values (filter
factors).
By default, this information is collected automatically by the system. You can manually control
the collection of statistics by manipulating the QDBFSTCCOL system value or by using the
iSeries Navigator graphical user interface (GUI). However, unlike indexes, statistics are not
maintained immediately as data in the tables changes.
There are cases were the optimizer advises the creation of statistics. The query shown in
Example 6-30 lists the optimizer advised statistics.
Example 6-30 Query optimizer advised statistics
SELECT
qqucnt, qvqtbl "Table", qvqlib "Schema",
qqc11 "Reason Stat Advised",
SUBSTR(qq1000,1,100) "Column name"
FROM mydbmon
WHERE qqrid=3015
ORDER BY 2,5;
167
Figure 6-26 Query optimizer advised statistics
򐂰 QQRID
Our selection criteria selects only 3015 rows, which contain information exclusive to SQE
advised statistics.
QQJFLD and QQUCNT are join fields that are required to uniquely identify the SQL
statement in the Database Monitor file.
These columns refer to the long SQL table name and the long SQL schema name. Use
QQTFN and QQTLN for short object names.
򐂰 QQC11
This column indicates the reason that the statistic was advised. This can happen only for
two reasons, where N indicates a new statistic and S indicates a stale statistic. A statistic
can become stale for several reasons; one of the most common reasons is that a base
physical table’s number of rows has changed by 15 percent.
򐂰 QQ1000
Column QQ1000 for row type 3015 contains the name of the column for which a statistic is
advised. There might be multiple recommendations for a single SQL query, with each row
containing a different column name in the QQ1000 column.
Keep in mind that column statistics are created in the background automatically by default for
all advised statistics. Therefore, in general, no manual action is required on your end to build
these statistics. The only exception is if the automatic statistics collection is turned off.
Although statistics provide a powerful mechanism for optimizing queries, do not
underestimate and disregard the importance of implementing a sound indexing strategy.
Well-defined indexes enable SQE to consistently provide efficient query optimization and
performance. Statistics cannot be used to access a table or sort the data during query
execution.
A good indexing strategy is both beneficial for providing statistics and mandatory for efficient
and fast query execution. Therefore, you should replace indexes with statistics only if the
168
indexes were created for the sole purpose of providing statistical information to the query
optimizer. In cases where an index provides fast and efficient access to a set of rows in a
table, DB2 Universal Database for iSeries continues to rely on its indexing technology to
provide statistics information and a data access method.
Since indexes are the premier way to improve query optimizer intelligence and data access
implementation choices, we look at the query in Example 6-31, which correlates SQE advised
statistics with any query optimizer recommended indexes.
Example 6-31 Correlation of SQE statistics and optimizer recommended indexes
WITH advisedIndexes AS (SELECT qqjfld, qqucnt, qqi2 "Nbr of Primary Keys",
SUBSTR(qqidxd, 1,100) "Keys Advised" FROM mydbmon
WHERE qqrid IN (3000, 3001, 3002) and qqidxa='Y') SELECT a.qqucnt, a.qqtfn "Table", a.qqtln
"Schema", CASE a.qqc11
WHEN 'N' THEN 'No Statistic Exists'
WHEN 'S' THEN 'Stale Statistic Exists'
ELSE 'Unknown'
END AS "Reason Stat Advised", SUBSTR(a.qq1000,1,100) "Column name", a.qqi2 "Statistics
Importance", a.qvc1000 "Statistics Identifier", b."Nbr of Primary Keys", b."Keys Advised"
FROM mydbmon a LEFT OUTER JOIN advisedIndexes b
on a.qqjfld = b.qqjfld and a.qqucnt = b.qqucnt WHERE qqrid=3015 ORDER BY a.qqucnt DESC;
Figure 6-27 Correlation of SQE statistic and optimizer recommended indexes
򐂰 QQIDXA
This column contains a flag of Y or N that indicates whether index was advised. We use
this information to filter out queries for which no index was recommended by the query
optimizer (QQIDXA = 'Y').
򐂰 QQIDXD
This column indicates columns for which an index was advised. This field lists primary
keys first followed by zero or more secondary keys. Secondary selection keys are less
likely to have a significant positive impact on a query’s performance.
򐂰 QQI2
This column indicates the number of primary keys contained in the QQIDXD field. For the
most benefit, build an index over the primary key fields.
169
򐂰 QQRID
We only focus on row types 3000, 3001, and 3002, which contain query optimizer index
suggestions.
򐂰 QQI2
This column indicates the importance of a statistic.
򐂰 QVC1000
This column contains the statistics identifier.
This query attempts to correlate the SQE advised statistics to the query optimizer index
suggestions. The idea is that we should attempt to build indexes for cases where an index can
be used by the query optimizer for the data access method.
If an index is solely used for statistical purposes, the advantage should be given to SQE
statistics due to their low maintenance overhead. An exception to this recommendation is if
statistics must be current at all times. The only way to accomplish this is by having an index
set with the *IMMED maintenance attribute.
6.3.14 Rows with retrieved or fetched details
Specifying *DETAIL in the TYPE parameter of the STRDBMON command indicates that detail
rows, as well as summary rows, must be collected for fetch operations. The same is true for
detailed the SQL Performance Monitor in iSeries Navigator.
The purpose of detailed 3019 row is for tuning non-SQL queries, those which do not generate
a QQ1000 row, such as OPNQRYF. For non-SQL queries, the only way to determine the
number of rows that are returned and the total time to return those rows is to collect detail
rows. While the detail row contains valuable information, it creates a slight performance
degradation for each block of rows returned. Therefore you must closely monitor its use. You
can use a detailed row for SQL analysis since the information it contains is also valuable in
the SQL environment.
A large number of physical I/O operations can indicate that a larger pool is necessary or that
SETOBJACC might be used to bring some of the data into main memory beforehand.
The query in Example 6-32 shows the most time consuming SQL statements.
Example 6-32 Most time consuming SQL statements
WITH retrieved AS (
SELECT
qqjfld, qqi3, qqi5
FROM mydbmon
WHERE
qqrid=3019)
SELECT
SUM(qqi6) "Total Time" , COUNT(*) "Nbr Times Run",
SUM(b.qqi3) "Sync DB Reads", SUM(b.qqi5) "ASync DB Reads", qq1000
FROM mydbmon a, retrieved b
WHERE
a.qqjfld=b.qqjfld AND qqrid=1000 AND qqucnt<>0 AND qqc21<>'MT'
170
Figure 6-28 Most time consuming SQL statements
򐂰 QQRID
In the common table expression retrieved, we select only 3019 rows, getting a subset of
data with detailed row information.
򐂰 QQJFLD
This column indicates the join column (unique per job).
򐂰 QQI3
This column indicates the number of synchronous database reads. We present a
cumulative count for each SQL statement.
򐂰 QQI5
This column indicates the number of asynchronous database reads. We present a
cumulative count for each SQL statement.
This column indicates the cumulative elapse time in microseconds for each individual
query. Since we use it as a cost indicator, we ordered the output in descending order
based on this value.
򐂰 QQ1000
This column contains the SQL statement.
򐂰 QQUCNT
This column contains the unique statement identifier. We use this value to exclude
nonunique statement identifiers.
򐂰 QQC21
This column contains the SQL request operation identifier. We use this value to exclude
continuation records from our analysis. Continuation records are used to display
statement text for statements that cannot fit into the single QQ1000 field.
This query gives us detailed information about the amount of reads that the longest running
SQL statements have performed.
171
Row 3019 contains other interesting statistics. We view them by running the query shown in
Example 6-33.
Example 6-33 Row 3019 statistics
WITH retrieved AS (
SELECT
qqjfld,qqi1,qqi2,qqi3,qqi5, qqi4,qqi6,qqi7,qqi8
FROM
mydbmon WHERE
qqrid=3019) SELECT
SUM(a.qqi6) "Total Time" , COUNT(*) "Nbr Times Run",
SUM(b.qqi1) "CPU time in milliseconds",
SUM(b.qqi2) "Clock Time in milliseconds", SUM(b.qqi3) "Sync DB Reads", SUM(b.qqi5)
"Async DB Reads",
SUM(b.qqi4) "Sync DB Writes", SUM(b.qqi6) "Async DB Writes",
SUM(b.qqi7)
"Number of rows returned",
SUM(b.qqi8) "Nbr of calls to get rows", qq1000 FROM mydbmon a, retrieved b WHERE
a.qqjfld=b.qqjfld AND qqrid=1000 AND qqucnt<>0 AND qqc21<>'MT' GROUP BY qq1000
Figure 6-29 shows the output of the query in Example 6-33 (we show two windows).
Figure 6-29 Row 3019 statistics
Note: In the following columns, we present cumulative values for each SQL statement that
generated the 3019 record. A statement is deemed unique as long as it can fit in the single
QQ1000 record.
򐂰 QQI1
This column indicates the CPU time in milliseconds.
򐂰 QQI2
This column indicates the clock time in milliseconds.
򐂰 QQI4
This column indicates the number of synchronous database writes.
172
򐂰 QQI6
This column indicates the number of asynchronous database writes.
򐂰 QQI7
This column indicates the number of rows returned.
򐂰 QQI8
This column indicates the number of calls to retrieve rows returned.
You can use QQI2 and QQI7 to calculate your row retrieval costs and then decide if using
SETOBJACC to bring the data to the memory pool beforehand will benefit the query in
question.
As mentioned earlier, row type 3019 is most useful when analyzing non-SQL queries, such as
OPNQRYF.
Now we look at the query in Example 6-34 that lists these queries and orders them based on
the elapsed time.
Example 6-34 Non-SQL queries
WITH retrieved AS (
SELECT
qqjfld,qqi2,qqi7
FROM mydbmon
WHERE
qqrid=3019 )
SELECT
(qqi1+b.qqi2) "Total Query Time", b.qqi7 "Number Rows Retrieved",
qqc101 "Open ID", qquser
FROM mydbmon a, retrieved b
WHERE a.qqjfld=b.qqjfld AND qqrid=3014
ORDER BY 1 DESC;
Figure 6-30 Non-SQL queries
This column indicates the time spent to open a cursor in milliseconds. It is added to the
fetch elapsed time and projected as the Total Query Time field.
The value in this column is our costing indicator. Any optimization efforts on non-SQL
queries should be focused on the most costly queries.
173
򐂰 QQC101
This column indicates the query open identifier. Non-SQL interfaces require job scooped
uniquely named identifiers at the open time. This identifier is contained in QQC101 field of
the row type 3014.
򐂰 QQUSER
This column indicates the job user name. Since most non-SQL interfaces run under the
real profile rather than the QUSER type profile, the QQUSER field is adequate for user
identification.
The intent of the query that we outlined here shows the most time consuming non-SQL
queries. Since there is no explicit flag in the Database Monitor data that differentiates
between SQL and non-SQL-based queries, we modify the query to try and exclude the open
IDs that are most likely SQL related as shown in Example 6-35.
Example 6-35 Non-SQL generated requests only
WITH qqq1000 AS (SELECT qqjfld,qqucnt FROM mydbmon
WHERE
qqrid=1000 AND qqucnt<>0 AND qqc21<>'MT'), retrieved AS (
SELECT
a.qqjfld,a.qqucnt,qqi2,qqi7
FROM mydbmon a EXCEPTION JOIN qqq1000 b ON
(a.qqjfld = b.qqjfld and a.qqucnt = b.qqucnt)
WHERE
qqrid=3019) SELECT
(qqi1+b.qqi2) "Total Query Time", b.qqi7 "Number Rows Retrieved",
qqc101 "Open ID", qquser "Job User", qqjob "Job Name",
qqjnum "Job Number" FROM mydbmon a INNER JOIN retrieved b ON
(a.qqjfld = b.qqjfld AND a.qqucnt = b.qqucnt)
WHERE
qqrid=3014 AND SUBSTR(qqc101,1,1) NOT IN (' ','*',x'00')
ORDER BY "Total Query Time" DESC;
Figure 6-31 Non-SQL generated requests only
򐂰 QQJOB
This column indicates the job name.
򐂰 QQJNUM
This column indicates the job number.
SQL-related rows generate record 1000, and non-SQL ones do not. We have taken
advantage of this fact to exclude SQL related rows from this query. Based on the start and
end time of Database Monitor collection, it is still possible for some SQL-related queries to be
included in our result set so we filter it further by checking the open ID’s starting character for
validity.
Even with this additional check, it is possible for some SQL-related rows to appear in our
result set. You might need to further customize the WHERE clause in the final select
statement.
174
6.3.15 Materialized query tables
Materialized query tables (MQTs), also referred to as automatic summary tables or
materialized views, can provide performance enhancements for queries. This enhancement
is done by precomputing and storing results of a query in the MQT. The database engine can
use these results instead of recomputing them for a user-specified query. The query optimizer
looks for any applicable MQTs and can choose to implement the query using a given MQT
provided this is a faster implementation choice.
MQTs are created using the SQL CREATE TABLE statement. Alternatively, the ALTER
TABLE statement might be used to convert an existing table into an MQT. The REFRESH
TABLE statement is used to recompute the results stored in the MQT. For user-maintained
MQTs, the MQTs might also be maintained by the user via INSERT, UPDATE, and DELETE
statements.
Support for creating and maintaining MQTs was shipped with the base V5R3 release of
i5/OS. The query optimizer support for recognizing and using MQTs is available with V5R3
i5/OS PTF SI17164 and the latest database group PTF SF99503 level 4.
For more information about MQTs, see the white paper The creation and use of materialized
query tables within IBM DB2 UDB for iSeries, which is available from the DB2 Universal
Database for iSeries Web site at:
http://www.ibm.com/iSeries/DB2
For the query shown here, you first build the MQT using SQL as shown in Example 6-36. For
this example, the employee table has roughly 2 million rows and the department table has five
rows.
Example 6-36 Creating the MQT
CREATE TABLE MQT1
AS (SELECT D.deptname, D.location, E.firstname, E.lastname, E.salary, E.comm,
E.bonus, E.job
FROM Department D, Employee E
WHERE D.deptno=E.workdept)
DATA INITIALLY IMMEDIATE REFRESH DEFERRED
ENABLE QUERY OPTIMIZATION
MAINTAINED BY USER
Then you run the query shown in Example 6-37, which returns information about employees
whose job is DESIGNER.
Example 6-37 Running SQL using MQT
SELECT D.deptname, D.location, E.firstname, E.lastname, E.salary+E.comm+E.bonus as
total_sal
FROM Department D, Employee E
WHERE D.deptno=E.workdept
AND E.job = 'DESIGNER'
In this query, the MQT matches part of the user’s query. The optimizer places the MQT in the
FROM clause and replaces the DEPARTMENT and EMPLOYEE tables. Any remaining
selection that is not done by the MQT query (M.job= 'DESIGNER') is done to remove the
extra rows. Then the result expression, M.salary+M.comm+M.bonus, is calculated.
175
Figure 6-32 shows the Visual Explain diagram of the query when using the MQT.
Figure 6-32 Visual Explain diagram for the MQT request
Notice the List of Materialized Query Tables in Figure 6-32. There are two tables, MQT2 and
MQT1. MQT2 is followed by a value of 4, and MQT1 is followed by a value of 0. The value of
4 means that the grouping specified in the MQT is not compatible with the grouping specified
in the query. The value of 0 means that the MQT was used in the query. These values are
retrieved from the qq1000 field of the 3030 record in SQL Performance Monitor data.
For a listing of all the values for QQ1000 field of the 3030 record, see DB2 Universal
Database for iSeries Database Performance and Query Optimization, which is available in the
iSeries Information Center on the Web at:
There are a number of ways to check whether an MQT is being used. Example 6-38 shows a
query that you can run to view the reasons why an MQT is not being used. You can also
check whether MQTs are replacing the existing table names in the query. You can check the
QQC13 column of either the 3000, 3001, or 3002 record of the SQL Performance Monitor
data. The 3000 record is for use when a table scan is done. The 3001 record is for use when
an index is used, and the 3002 record is for when an index is created.
176
To look at the MQTs on your system and see if they are being used, run the following Display
File Description (DSPFD) command and SQL Performance Monitor data to check. First you
must determine which tables are MQTs. Create a file with the following command:
DSPFD FILE(library or *ALLUSR/*ALL) TYPE(*ATR) OUTPUT(*OUTFILE) FILEATR(*PF)
OUTFILE(library/MQTS)
Then to see which MQTs are being used or are not in queries, run the query shown in
Example 6-38 against the file created from the DSPFD command, using SQL performance
data.
Example 6-38 Query to find MQTs
with MQTFILES as (select phfile, phlib from library/MQTS
where phsqlt = 'M')
select count(*) as Times_Used,
a.phfile as MQT_file,
a.phlib as MQT_library,
b.qqc13 as Used_in_Query
from MQTFILES a left outer join library/dbmondata b
on a.phfile = b.qqtfn and a.phlib = b.qqtln
group by a.phfile, a.phlib, b.qqc13
order by b.qqc13
Next, to determine why MQTs are not being used, run the query shown in Example 6-39.
Example 6-39 Reason why MQTs are not used
SELECT substr(a.qq1000,1,100) AS MQT_RC,b.qq1000
FROM library/dbmondata a,
library/dbmondata b where a.qqrid = 3030 and b.qqrid = 1000 and a.qqjfld
= b.qqjfld and a.qq1000 <> '0' and b.qqc11 in ('S','U','I','D') and
b.qqc21 NOT IN ('MT', 'CL', 'HC', 'FE')
GROUP BY B.QQ1000, A.QQ1000
177
178
7
Chapter 7.
Using Collection Services data to
identify jobs using system
resources
This chapter explains how to find jobs using CPU or I/O with Collection Services data and
how you can integrate it with the Database Monitor data. This chapter begins by describes
how to start Collection Services. Then it guides you in using Collection Services data to find
jobs using CPU, to find jobs with high disk I/O counts, and to use it in conjunction with the
SQL Performance Monitors.
179
7.1 Collection Services and Database Monitor data
Collection Services allows you to gather performance data with little or no observable impact
on system performance. Its data is analyzed using the IBM Performance Tools for iSeries
licensed program (5722PT1) or other performance report applications, iSeries Navigator
monitors, and the graph history function. If you prefer to view real-time performance data,
system monitors provide an easy-to-use graphical interface for monitoring system
performance. For more information about iSeries Navigator Monitors, see “iSeries Navigator
monitors” in iSeries Performance Version 5 Release 3 on the Web at:
http://publib.boulder.ibm.com/infocenter/iseries/v5r3/ic2924/info/rzahx/rzahx.pdf
Collection Services collects data that identifies the relative amount of system resource used
by different areas of your system. While you are required to use Performance Tools for
iSeries, it is not required to gather the data.
In most of the analysis when looking for job-level information in this chapter, we use the
Component Report. The Component Report uses Collection Services data to provide
information about the same components of system performance as a System Report, but at a
greater level of detail. The Component Report helps you find the jobs that are consuming high
amounts of system resources, such as CPU, disk, and so on.
Note: While running Collection Services, you should also run Database Monitor on all the
jobs. After we identify the job information of the job or jobs that consume high amounts of
system resources, such as CPU, disk and so on, we can the query the Database Monitor
data to find the jobs as explained in Chapter 6, “Querying the performance data of the
Database Monitor” on page 133.
7.1.1 Starting Collection Services
To analyze Collection Services data, you must first start Collection Services, which you can
do in one of the following ways:
򐂰
򐂰
򐂰
򐂰
From the PERFORM menu, using GO PERFORM
From iSeries Navigator
Using Performance Management APIs
In V5R3, using the Start Performance Collection (STRPFRCOL) CL command
Note: Starting Collection Services from the PERFORM menu or from iSeries Navigator
requires that you install Performance Tools for iSeries, 5722PT1.
In the following sections, we explain the details for each of the methods to start Collection
Services to gather data.
180
PERFORM menu
To start Collection Services, from the PEFORM menu:
1. Enter the following command:
GO PERFORM
2. On the IBM Performance Tools for iSeries display, type option 2 (Collect performance
data).
3. On the Collect Performance Data display, select option 2 (Configure Performance
Collection).
4. The Configure Perf Collection (CFGPFRCOL) display is shown. In setting the
configuration, you should consider your system resources, for example, whether you have
enough storage to handle the data being gathered. You must consider the following two
parameters:
– Default interval: This parameter indicates the sample interval time in minutes to collect
data.
– Collection retention period: This parameter indicates how long the management
collection object (*MGTCOL) should be retained on the system.
Set these parameters based on your system resources and the problem that you are
trying to capture. For example, if you have a performance problem that occurs
intermittently over a week, you might set the interval time to 15 minutes and the collection
retention period to 7 days or longer if you want comparison data, but only if you have the
resources to maintain the size of the *MGTCOL objects.
After you set the configuration, press Enter.
5. Type option 1 (Start Performance Collection). Specify *CFG for the collection profile.
For more information, see iSeries Performance Version 5 Release 3, which you can find on
the Web at:
7.1.2 From iSeries Navigator
In iSeries Navigator, select your system name → Configuration and Service. Right-click
Collection Services and select Start Performance Collection. In the Start Collection
Services window (Figure 7-1), verify the configuration settings and click OK to start Collection
Services.
Chapter 7. Using Collection Services data to identify jobs using system resources
181
Figure 7-1 Configuring and starting Collection Services in iSeries Navigator
7.1.3 Using Performance Management APIs
You can use Collector APIs to start collecting performance data. The APIs do not require you
to have Performance Tools for iSeries installed. To start Collection Services, you can use the
following command:
CALL QYPSSTRC PARM('*PFR
' '*STANDARDP' X'00000000')
For more information about the parameters of the API, see “Collector APIs” in iSeries
Performance Management APIs Version 5 Release 3 on the Web at:
http://publib.boulder.ibm.com/infocenter/iseries/v5r3/ic2924/info/apis/perfmg.pdf
7.1.4 V5R3 STRPFRCOL command
You can also start V5R3 of Collection Services by using the STRPFRCOL CL command.
Then to configure Collection Services, you use the Configure Performance Collection
(CFGPFRCOL) CL command.
For more information about the Collection Services CL commands, see iSeries Performance
Version 5 Release 3 on the Web at:
182
7.2 Using Collection Services data to find jobs using CPU
You can use Collection Services data to look for jobs using CPU. In this section, we explain
how to find jobs that are using CPU by using the following methods:
򐂰 Component Report from the PERFORM menu
򐂰 iSeries Navigator Graph History
򐂰 Management Central System Monitors
Note: When the job information is found, you can use the job information to query the
Database Monitor table as explained in Chapter 6, “Querying the performance data of the
Database Monitor” on page 133.
7.2.1 Finding jobs using CPU with the Component Report
To use the PERFORM menu of IBM Performance Tools for iSeries to gather the Component
Report, you must install Performance Tools, product 5722PT1. To access the PERFORM
menu, enter the following command:
GO PERFORM
If you want to interactively examine the data, select option 7 (Display performance data). If
you choose this option, you must keep in mind your system resources. This job runs
interactively and uses CPU and I/O resources. Or you can select option 3 (Print performance
report), which submit a job to batch.
After you select an option, you then select the member that you want to investigate, based on
the date and time shown. If no members are shown and you have started Collection Services,
run the Create Performance Data (CRTPFRDTA) command against the *MGTCOL object that
contains your data or create the files in iSeries Navigator.
򐂰 To create the files using the CRTPFRDTA CL command, find the *MGTCOL object to
create the files from by entering the following command:
WRKOBJ OBJ(qmpgdata/*ALL) OBJTYPE(*MGTCOL)
The attribute of the *MGTCOL object must be *PFR. In the WRKOBJ command shown,
you replace qmpgdata with the library where you keep your performance data. After the
*MGTCOL object is found, you run the following CL command to create the database files:
CRTPFRDTA FROMMGTCOL(library/mgtcolname)
183
򐂰 To create the files using iSeries Navigator, select your system name → Configuration
and Service. Right-click Collection Services. Then in the right pane, you see a list of the
*MGTCOL objects. Right-click the collection name, which is the *MGTCOL object, and
select Create Database Files Now as shown in Figure 7-2.
For more information, see iSeries Performance Version 5 Release 3 on the Web at:
Figure 7-2 Creating files from *MGTCOL objects in iSeries Navigator
The following sections explain how to find the jobs that are using CPU by using either the
option to print performance report or the option to display performance data.
Using option 3: Print performance report
When you select option 3, you must specify the library in which you keep your performance
data. In most cases, the library is QMPGDATA. You can page up and down until you see a
time frame that you want. Type 2 next to the member that you want to use to get a Component
Report.
In the next Select Sections for Report display (Figure 7-3), you can press F6 to print the entire
report to select all the sections to be in your report. Since you must often review all the
performance data at the time of high CPU usage, we recommend that you use F6. In the
display shown in Figure 7-3, you see that we choose option 1 to select Job Workload Activity
so that we can only see the jobs that are using CPU.
184
Select Sections for Report
Member . . . . . . . . . . . :
Type options, press Enter.
1=Select
Option
1
Q075000002
Press F6 to print entire report.
Section
Component Interval Activity
Job Workload Activity
Storage Pool Activity
Disk Activity
IOP Utilizations
Local Work Stations
Remote Work Stations
Exception Occurrence
Data Base Journaling Summary
TCP/IP Activity
HTTP Server Activity
Bottom
F3=Exit
F6=Print entire report
F12=Cancel
Figure 7-3 Select Sections for Report display
In the Select Categories for Report display (Figure 7-4), we type option 1 to select Time
interval. Time interval is usually the best option if you don’t know any other information. If you
want more information, you can press F6 to print the entire report, but it is best to narrow the
information down to a time frame in which you are interested.
Select Categories for Report
Member . . . . . . . . . . . :
1=Select
Q075000002
Press F6 to print entire report.
Option
1
Category
Time interval
Job
Job type
Job run priority
User ID
Subsystem
Pool
Communications line
Control unit
Functional area
F3=Exit
F6=Print entire report
Bottom
F12=Cancel
Figure 7-4 Select Categories for Report display
185
If you selected Time interval on the previous display, you see the Select Time Intervals display
(Figure 7-5). You can select intervals that you want to look at based on CPU utilization. In this
example, we select two time frames of interest by typing 1 in the Opt column. These time
frames were chosen because the concern is the sudden CPU growth.
Select Time Intervals
Library . . . . . :
QMPGDATA
1=Select
O
p
Transaction
t Date
Time
Count Resp
03/15 07:45
0 .00
03/15 08:00
0 .00
03/15 08:15
2 .00
03/15 08:30
0 .00
03/15 08:45
8 .00
03/15 09:00
0 .00
03/15 09:15
0 .00
03/15 09:30
104 .24
1 03/15 09:45
86 .59
1 03/15 10:00
374 .14
F3=Exit
F13=Sort (date/time)
Performance data . . . . . :
Int
-CPU Util-- Feat
Tot Int Bch Util
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2
2
0
23
10 10
0 127
17
9
8 124
High
--Util-Dsk Unit
1 0001
-- ---1 0003
1 0001
1 0001
1 0002
1 0005
4 0001
6 0001
11 0001
F5=Refresh
F14=Sort (count)
Q074000002
Pool
-Fault/SecMch User ID
0
0 02
0
0 02
0
0 02
0
0 02
0
0 02
0
0 02
0
0 02
0
0 03
0
0 03
0
0 03
Excp
39
39
120
341
96
168
505
163
55
1710
More...
F12=Cancel
F24=More keys
Figure 7-5 Select Time Intervals display
In the Specify Report Options display (Figure 7-6), you specify any report title that you want.
In this example, we specify CPU Report. You press Enter, and a job is submitted to batch.
Specify Report Options
Type choices, press Enter.
Report title . . . .
CPU Report
Report detail . . .
*JOB
*JOB, *THREAD
Job description . .
Library . . . . .
QPFRJOBD
*LIBL
Name, *NONE
Name, *LIBL, *CURLIB
F3=Exit
F12=Cancel
Figure 7-6 Specify Report Options panel
You then return to the Print Performance Report - Sample data display and see the following
message at the bottom of the display:
Job 003675/PCHIDESTER/PRTCPTRPT submitted to job queue QBATCH in library ...
This message gives you the submitted job for your report. You can find your report by entering
the following command:
WRKSBMJOB *JOB
186
In the Work with Submitted Jobs display, your submitted job is called PRTCPTRPT. When the
PRTCPTRPT job is in OUTQ status, select option 8 to view the spool file. The report is in the
spool file QPPTCPTR. Type 5 on the line that precedes the QPPTCPTR file.
In Figure 7-7, you see part of the Component Report showing the jobs that are using CPU. In
this example, there are three jobs of concern as highlighted in bold.
File . . . . . :
QPPTCPTR
Control . . . . .
Find . . . . . .
*...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8....+....9....+....0....
Virtual Processors:
2
Processor Units :
2.0
T
P
DB
Job
User Name/ Job
y
t
CPU Cpb
Tns
-------- Disk I/O -------Name
Thread
Number p Pl y
Util Util
Tns
/Hour
Rsp
Sync
Async Logical
---------- ---------- ------ -- -- -- ----- ----- ----- --------- ------- -------- -------- -------- QTVDEVICE QTCP
003297 B 02 20
.00
.0 0
0
.000
0
0
0
QTVDEVICE QTCP
003298 B 02 20
.00
.0
0
0
.000
0
0
0
QTVDEVICE QTCP
003299 B 02 20
.00
.0
0
0
.000
60
40
0
QTVDEVICE QTCP
003300 B 02 20
.00
.0
0
0
.000
0
0
0
QTVDEVICE QTCP
003301 B 02 20
.00
.0
0
0
.000
0
0
0
QTVDEVICE QTCP
003302 B 02 20
.00
.0
0
0
.000
0
0
0
QUERY
PCHIDESTER 003460 B 02 50 6.64 1.7
0
0
.000
5602
32789
5161
QUERY2
PCHIDESTER 003461 B 02 50 7.72 2.0
0
0
.000
5573
24476
5161
QUERY3
PCHIDESTER 003462 B 02 50 7.82 1.9
0
0
.000
5529
22520
5161
QUSRWRK
QSYS
003241 M 02 00
.00
.0
0
0
.000
0
0
0
QYPSJSVR
QYPSJSVR
003314 B 02 16
.09
.0
0
0
.000
9
0
0
QYPSPFRCOL QSYS
003354 B 02 01
.00
.0
0
0
.000
0
39
0
QYUSCMCRMD QSYS
003312 B 02 50
.00
.0
0
0
.000
1
0
0
Q1PSCH
QPM400
003281 B 02 50
.00
.0
0
0
.000
0
0
180
F3=Exit
F12=Cancel
F19=Left
F20=Right
F24=More keys
Figure 7-7 Partial print of Jobs using CPU
The disadvantage of using the printed report versus displaying the data interactively is that
the printed Component Report is sorted by the job name, not by CPU. Displaying the data
interactively allows you to sort on CPU.
Now that we have the jobs that are using the majority of CPU, we can now look at the
Database Monitor data that was running at the same time to investigate what the jobs were
doing. Chapter 6, “Querying the performance data of the Database Monitor” on page 133,
gives examples on how to investigate the SQL, if any, that the jobs were running based on the
job name, job user, and job number.
Option 7: Display performance data
If you select option 7, you must specify the library in which you keep your performance data.
In most cases, the library is QMPGDATA. Next you select the member that you want to
investigate, based on the date and time shown. You do this by typing 1 next to the member.
After you select the member, you can page up and down until you see a time frame you are
looking for or until you see a CPU Utilization that concerns you. Then you type 1 next to the
interval or intervals that you want to examine.
187
After you select the intervals, you see the Display Performance Data display (Figure 7-8).
Display Performance Data
Member . . . . . . . .
Library . . . . . .
Elapsed time
System . . .
Start date .
Start time .
Partition ID
QPFRADJ . .
QDYNPTYSCD .
QDYNPTYADJ .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
:
:
:
:
:
:
:
:
Q074000002
QMPGDATA
00:15:00
RCHASM05
03/15/05
00:00:02
000
0
1
1
CPU utilization (interactive) .
CPU utilization (other) . . . .
Interactive Feature Utilization
Time exceeding Int CPU Threshold
Job count . . . . . . . . . . .
Transaction count . . . . . . .
F3=Exit
F12=Cancel
F4 for list
Version . . . . .
Release . . . . .
Model . . . . . .
Serial number . .
Feature Code . . .
Int Threshold . .
Virtual Processors
Processor Units .
. .
. .
. .
(in
. .
. .
. . . . .
. . . . .
. . . . .
seconds)
. . . . .
. . . . .
F4=Prompt F5=Refresh
F24=More keys
.
.
.
.
.
.
:
:
:
:
:
:
:
:
:
:
:
:
:
:
5
3.0
270
10-5HZ4M
23F5-2434-1520
0.51 %
2
2.00
10.10
.54
127.48
85882
22
86
F6=Display all jobs
More...
F10=Command entry
Figure 7-8 Display Performance Data display
From this display, press F6 to view all jobs. Then you see a listing of the jobs that were
running during the interval as shown in Figure 7-9. You can use F19 to sort by CPU.
Display Jobs
Elapsed time . . :
00:15:00
Member . . . . . :
Library . . . . :
Y074000002
QMPGDATA
5=Display job detail 6=Wait detail
Option
Job
QPADEV0001
QPADEV0003
CFINT01
QPADEV0005
QYPSJSVR
CFINT02
QPADEV0004
QINTER
CRTPFRDTA
QYPSPFRCOL
F3=Exit
F19=Sort by CPU
Job
User
Number Type
PCHIDESTER 003456 PTH
LIC
QYPSJSVR
003314 BCH
LIC
QSYS
003247 SBS
QSYS
003425 BCH
QSYS
003354 BCH
F12=Cancel
F15=Sort by job
F24=More keys
CPU
Util
6.62
3.30
.31
.13
.10
.09
.04
.00
.00
.00
Avg
Rsp
.0
.0
.0
1.6
.0
.0
.2
.0
.0
.0
F16=Sort by job type
Figure 7-9 Listing of jobs using CPU
188
Tns
Count
0
0
0
20
0
0
66
0
0
0
Disk
I/O
69347
40359
0
2188
5
0
500
14
194
20
More...
In this example, it appears as though there are two jobs that are using the majority of CPU. If
this is unusual and Database Monitor data was gathered during this interval, you can run a
query based on the job name, job user, and job number as shown in Chapter 6, “Querying the
performance data of the Database Monitor” on page 133. The query helps to isolate the SQL,
if any, that these jobs were running during this interval.
7.2.2 Finding jobs using CPU with iSeries Navigator Graph History
Graph history provides a graphical view of performance data collected over days, weeks,
months, or years with Collection Services. You do not need to have a system monitor running
to view performance data. As long as you use Collection Services to collect data, you can
view the Graph History window. To access the graph history in iSeries Navigator, select the
system name → Configuration and Service. Right-click Collection Services and select
Graph History as shown in Figure 7-10.
Figure 7-10 Selecting Graph History
189
Then the Graph History window opens. When you click the drop-down list for Metric, you see
several options to help you find the jobs using CPU as shown in Figure 7-11. In this example,
we select the option CPU Utilization (Average). You also need to specify the time frame for
which you want to see graph history.
Figure 7-11 Graph History metric options
After you launch a graph history, a window opens that shows a series of graphed collection
points. These collection points on the graph line are identified by three different graphics that
correspond to the three levels of data that are available:
򐂰 A square collection point represents data that includes both the detailed information and
properties information.
򐂰 A triangular collection point represents summarized data that contains detailed
information.
򐂰 A circular collection point represents data that contains no detailed information or
properties information.
Figure 7-12 shows information after a point on the graph was selected to show the graph in
the upper right corner giving the job names. You can select any of the bars on the graph to
view information about the job in the box underneath it. In the example shown, the bar turns
black after being selected, and the first three lines of information in the box underneath the
graph are the job name, user name, and job number. This information is required to run the
queries in Chapter 6, “Querying the performance data of the Database Monitor” on page 133,
to indicate whether the jobs were running SQL, and if so, the SQL statement that was run.
190
Figure 7-12 Displaying job using CPU with Graph History data
For more information about using Graph History see iSeries Performance Version 5
Release 3 on the Web at:
7.2.3 Finding jobs using CPU with Management Central System Monitors
The system monitors display the data stored in the collection objects that are generated and
maintained by Collection Services. The system monitors display data as it is collected, for up
to one hour. To view longer periods of data, use Graph history as explained in 7.2.2, “Finding
jobs using CPU with iSeries Navigator Graph History” on page 189. You can change the
frequency of the data collection in the monitor properties, which overrides the settings in
Collection Services.
To set up a system monitor, you must define the metrics that you want to track and the action
that the monitor should take when the metrics reach the specified levels. To define a system
monitor that looks for jobs using CPU, complete the following steps:
1. In iSeries Navigator, expand Management Central → Monitors. Right-click System
Monitor and select New Monitor.
2. On the General page, enter a name and description for the monitor.
3. On the Metrics page, and enter the following values:
a. From the list of Available Metrics, select CPU Utilization (Average) and click Add.
CPU Utilization Basic (Average) is now listed under Metrics to monitor, and the bottom
portion of the window shows the settings for this metric.
191
b. Add two other Available Metrics: CPU Utilization (Interactive Feature) and CPU
Utilization (Database Capability).
c. For Collection interval, specify how often you want to collect this data. This value
overrides the Collection Services setting. For this example, we specify 30 seconds.
d. To change the scale for the vertical axis of the monitor’s graph for this metric, change
the Maximum graphing value. To change the scale for the horizontal axis of the graph
for this metric, change the value for Display time.
4. On the Systems and Groups page, select your systems and groups. You can click Browse
to select the endpoint system on which you want to run the monitor.
5. Expand Management Central → Monitors and click System. You should now see the
monitor displayed with a status of Stopped.
6. To start the monitor, right-click it and select Start.
7. To view the graph data, double-click the monitor that you created. Click any square in the
graph to see a bar graph in the upper right corner that shows the job names sorted by
CPU. To find more job details, click any of the bars on the graph as shown in Figure 7-13.
Figure 7-13 System Monitor showing job using CPU
192
The box below the graph shows the fully qualified job name, consisting of job name, user
name, and job number. You can use this information to query a Database Monitor file
collecting during this time frame as shown in Chapter 6, “Querying the performance data of
the Database Monitor” on page 133.
For more information about using Graph History see iSeries Performance Version 5
Release 3 on the Web at:
7.3 Using Collection Services data to find jobs with high disk
I/O counts
You can use Collection Services data to look for jobs with high disk I/O counts. To find the
jobs with high disk I/O, you are required to have Performance Tools installed. To start looking
for jobs that have a high I/O count, use the Component Report from the PERFORM menu. To
access the PERFORM menu, enter the following command:
GO PERFORM
If you want to interactively view the data, select option 7 (Display performance data). Keep in
mind your system resources when you choose this option. This job runs interactively and
uses CPU and I/O resources. Or you can choose option 3 (Print performance report), which
submits a job to batch.
The following sections explain how to find the jobs with high I/O counts either interactively
(option 7) or in batch (option 3).
Option 3: Print performance report
When you select option 3, you must specify the library in which you keep your performance
data. In most cases, the library is QMPGDATA. You can page up and down until you see a
time frame that you want. Type 2 next to the member that you want to use to get a Component
Report.
If no members are found and you have started Collection Services, you must create the
database files that are needed. See the first bullet point in 7.2.1, “Finding jobs using CPU with
the Component Report” on page 183, for information about creating the files.
In the next Select Sections for Report display (Figure 7-3 on page 185), you can press F6
(Print entire report) to select all the sections to be in your report. Since you must often review
the storage pool activity and disk activity, we recommend that you use F6. In the display
shown in Figure 7-3, you see that we choose option 1 to select Job Workload Activity to find
the jobs with high I/O counts.
In the Select Categories for Report display (Figure 7-4 on page 185), we type option 1 to
select Time interval. Time interval is usually the best option if you don’t know any other
information. If you want more information, you can press F6 to print the entire report, but it is
best to narrow the information down to a time frame in which you are interested.
If you selected Time interval on the previous display, you see the Select Time Intervals display
(Figure 7-5). You can select intervals that you want to look at based on High Disk Utilization or
a time frame.
In the Specify Report Options display (Figure 7-6 on page 186), you specify any report title
that you want. You press Enter, and a job is submitted to batch.
193
You then return to the Print Performance Report - Sample data display and see the following
message at the bottom of the display:
Job 003675/PCHIDESTER/PRTCPTRPT submitted to job queue QBATCH in library ...
This message gives you the submitted job for your report. You can find your report by entering
the following command:
WRKSBMJOB *JOB
In the Work with Submitted Jobs display, your submitted job is called PRTCPTRPT. When the
PRTCPTRPT job is in OUTQ status, select option 8 to view the spool file. The report is in
spool file QPPTCPTR. Type 5 on the line that precedes the QPPTCPTR file.
Figure 7-14 shows a job that has a high I/O count in comparison to other jobs in the report.
The disadvantage of using the printed report versus displaying the data interactively is that
the printed Component Report is sorted by the job name, not by disk I/O. Displaying the data
interactively allows you to sort on disk I/O. Now that we have isolated a job with relatively high
disk I/O, we can look at the Database Monitor data that was running at the same time to
investigate what the job was doing. Chapter 6, “Querying the performance data of the
Database Monitor” on page 133, gives examples of how to investigate what SQL, if any, that
the job was running based on the job name, job user, and job number.
Display Spooled File
File . . . . . :
Control . . . . .
Find . . . . . .
QPPTCPTR
*...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8....+....9....+....0....
Library . . : QMPGDATA System name . . :RCHASM05
Version/Release :
5/ 3.0 Stopped
Partition ID : 000
Feature Code . :23F5-2434-1520
Int Threshold . :
.51 %
Virtual Processors: 2
Processor Units : 2.0
T
P
DB
Job
User Name/ Job
y
t
CPU Cpb
Tns
-------- Disk I/O -------Name
Thread
Number p Pl y
Util Util Tns
/Hour
Rsp
Sync
Async Logical
---------- ---------- ------ -- -- -- ----- ----- ------- --------- ------- -------- -------- -------- +
QINTER
QSYS
003247 M 02 00
.00
.0
0
0
.000
13
1
0
QPADEV0003 DSQUIRES
003789 P 03 20 28.36 24.0
236
944
.004
4751
267364
847174
QPADEV0004 DSQUIRES
003836 P 03 20
.05
.0
168
828
.051
186
80
63
QPADEV0005 PCHIDESTER 003782 P 03 20
.00
.0
28
112
.051
54
82
789
F3=Exit
F12=Cancel
F19=Left
F20=Right
F24=More keys
Figure 7-14 Example of job having relative high I/O count
Option 7: Display performance data
If you select option 7, you must specify the library in which you keep your performance data.
In most cases, the library is QMPGDATA. Next you select the member that you want to
investigate, based on the date and time shown. You do this by typing 1 next to the member.
If no members are found and you started Collection Services, you must create the database
files needed. See 7.2.1, “Finding jobs using CPU with the Component Report” on page 183,
for information about how to create the files.
After you select the member, you can page up and down until you see a time frame you are
looking for or until you see a high disk utilization that concerns you. Then you type 1 next to
the interval or intervals that you want to examine.
194
After you select the intervals, you see the Display Performance Data display (Figure 7-8 on
page 188). From this display, press F6 to view all jobs. You can use F22 to sort by disk I/O as
shown in Figure 7-15. If the disk activity is unusually high and Database Monitor data was
gathered during this interval, then the Database Monitor data can be queried as explained in
Chapter 6, “Querying the performance data of the Database Monitor” on page 133, to isolate
what the job is doing.
Display Jobs
Elapsed time . . :
00:15:00
Member . . . . . :
Library . . . . :
Q075000002
QMPGDATA
5=Display job detail 6=Wait detail
Option
Job
QPADEV0003
QUERY
BHAUSER_04
QUERY
PRTCPTRPT
QZDASOINIT
QSYSCOMM1
QZDASOINIT
QDBSRVXR2
QDBFSTCCOL
F3=Exit
F19=Sort by CPU
User
DSQUIRES
DSQUIRES
BHAUSER
DSQUIRES
PCHIDESTER
QUSER
QSYS
QUSER
QSYS
QSYS
Number
003789
003834
003760
003835
003837
003738
003220
003822
003219
003221
Job
Type
PTH
BCH
PTH
BCH
BCH
BCH
SYS
BCH
SYS
SYS
F12=Cancel
F15=Sort by job
F24=More keys
CPU
Util
28.36
2.95
1.11
2.46
.16
.26
.00
.14
.00
.02
Tns
Count
236
0
126
0
0
0
0
0
0
0
Avg
Rsp
.0
.0
.9
.0
.0
.0
.0
.0
.0
.0
Disk
I/O
272115
42815
40255
38044
1748
1274
1113
1066
843
667
More...
F16=Sort by job type
Figure 7-15 Job having a high disk I/O count
You can find additional information about Collection Services data and using the tools in
“Performance Tools Reports” in the Information Center at:
http://publib.boulder.ibm.com/infocenter/iseries/v5r3/index.jsp?topic=/rzahx/
rzahxreportperftools.htm
Also refer to iSeries Performance Version 5 Release 3, which is available on the Web at:
195
196
8
Chapter 8.
Analyzing database performance
data with Visual Explain
The launch of Visual Explain with iSeries Navigator Version 4, Release 5, Modification 0
(V4R5M0) in DB2 Universal Database for iSeries was of great interest to database
administrators working in an iSeries server environment. This feature has been described as
a quantum leap forward in database tuning for query optimization. Visual Explain provides an
easy-to-understand graphical interface that represents the optimizer implementation of the
query.
For the first time, you can see, in graphic detail, how the optimizer has implemented the
query. You can even see all of the facts and figures that the optimizer used to make its
decisions. Best of all, the information is presented in one place, in color, with easy-to-follow
displays. There is no more jumping between multiple windows, trying to determine what is
happening. Even better, if you currently have iSeries Navigator, you already have Visual
Explain.
With all of this in mind, is such a richly featured product complicated to use? As long as you
are familiar with database tuning, you will enjoy using Visual Explain and want to learn more.
This chapter answers the following questions:
򐂰
򐂰
򐂰
򐂰
򐂰
Where do I find Visual Explain?
How do I use it?
What can it be used for?
Will it tune my Structured Query Language (SQL) queries?
How can we integrate Visual Explain with the Database Monitors?
197
8.1 What is Visual Explain
Visual Explain provides a graphical representation of the optimizer implementation of a query
request. The query request is broken down into individual components with icons that
represent each unique component. Visual Explain also includes information about the
database objects that are considered and chosen by the query optimizer. Visual Explain’s
detailed representation of the query implementation makes it easier to understand where the
greatest cost is being incurred.
Visual Explain shows the job run environment details and the levels of database parallelism
that were used to process the query. It also shows the access plan in diagram form, which
allows you to zoom to any part of the diagram for further details.
If query performance is an issue, Visual Explain provides information that can help you to
determine whether you need to:
򐂰 Rewrite or alter the SQL statement
򐂰 Change the query attributes or environment settings
򐂰 Create new indexes
Best of all, you do not have to run the query to find this information. Visual Explain has a
modeling option that allows you to explain the query without running it. That means that you
can try any of the changes suggested and see how they are likely to work, before you decide
whether to implement them.
You can also use Visual Explain to:
򐂰 View the statistics that were used at the time of optimization
򐂰 Determine whether an index was used to access a table
If an index was not used, Visual Explain can help you determine which columns might
benefit from being indexed.
򐂰 View the effects of performing various tuning techniques by comparing the before and
after versions of the query graph
򐂰 Obtain information about each operation in the query graph, including the total estimated
cost and number of rows retrieved
򐂰 View the debug messages issued by the query optimizer during the query execution
Visual Explain is an advanced tool to assist you with the task of enhancing query
performance, although it does not actually do this task for you. You still need to understand
the process of query optimization and the different access plans that you can implement.
Visual Explain is a perfect match with the Database Monitors.
8.2 Finding Visual Explain
Visual Explain is a component of iSeries Navigator and is available under the Databases icon.
To locate the Databases icon, you must establish a session on your selected iSeries server
using the iSeries Navigator icon.
From the SQL Script Center, you can access to Visual Explain directly, either from the menu
or from the toolbar as explained in 8.3.1, “The SQL Script Center” on page 199.
198
Another way to access Visual Explain is through the SQL Performance Monitors. SQL
Performance Monitor is used to create Database Monitor data and to analyze the monitor
data with predefined reports.
Visual Explain works with the monitor data that is collected by SQL Performance Monitor on
that system or by the Start Database Monitor (STRDBMON) command. Visual Explain can
also analyze Database Monitor data that is collected on other systems after that data is
restored on the iSeries server.
8.3 Using Visual Explain with the SQL Script Center
The Run SQL Script window (SQL Script Center) provides a direct route to Visual Explain.
The window is used to enter, validate, and execute SQL commands and scripts and to provide
an interface with i5/OS through the use of CL commands.
8.3.1 The SQL Script Center
To access the SQL Script Center, in iSeries Navigator, expand Databases. Then select any
database, right-click, and select Run SQL Scripts. The Run SQL Script window (Figure 8-1)
opens with the toolbar. Reading from left to right, there are icons to create, open, and save
SQL scripts, followed by icons to cut, copy, paste, and insert generated SQL (from V5R1)
statements within scripts.
The hour glass icons indicate to run the statements in the Run SQL Scripts window. From left
to right, they run all of the statements, run all of the statements from the cursor to the end
(From Selected), or run the single statement identified by the cursor position (Selected).
To the right of the hour glasses is a Stop button, which is in the color red when a run is in
progress.
There are two Visual Explain icons in the colors blue and green. The left Visual Explain icon
(blue) helps to explain the SQL statement. The right Visual Explain icon (green) enables you
to run and explain the SQL statement.
Figure 8-1 Toolbar from Run SQL Scripts
Both of these options are also available from the Visual Explain menu (Figure 8-2). You might
choose either option to start Visual Explain.
Figure 8-2 SQL Script Center Visual Explain options
The final icon in the toolbar is the Print icon.
Chapter 8. Analyzing database performance data with Visual Explain
199
You can use SQL Performance Monitors to record SQL statements that are explainable by
Visual Explain. We recommend that you obtain access via the SQL Performance Monitors
icon, because it provides the full list of monitors.
8.3.2 Explain Only
The Visual Explain Only option (Ctrl + E or the blue toolbar icon) submits the query request to
the optimizer and provides a visual explanation of the SQL statement and the access plan
that will be used when executing the statement. In addition, it provides a detailed analysis of
the results through a series of attributes and values associated with each of the icons. It does
not actually run the SQL statement.
To optimize an SQL statement, the optimizer validates the statement, gathers statistics about
the SQL statement, and creates an access plan. When you choose the Visual Explain Only
option, the optimizer processes the query statement internally with the query time limit set to
zero. Therefore, it proceeds through the full validation, optimization, and creation of an access
plan and then reports the results in a graphical display.
Note: When you choose Visual Explain Only, Visual Explain might not be able to explain
such complex queries as hash join, temp join results, and so on. In this case, you must
choose Run and Explain for the SQL statements to see the graphical representation.
8.3.3 Run and Explain
The Run and Explain option (Ctrl + U or the green toolbar icon) also submits the query
request to the optimizer. It provides a visual explanation of the SQL statement and the access
plan used when executing the statement. It provides a detailed analysis of the results through
a series of attributes and values associated with each of the icons.
However, it does not set the query time limit to zero and, therefore, continues with the
execution of the query. This leads to the display of a results window in addition to the Visual
Explain graphics.
Notes:
򐂰 Visual Explain might show a representation that is different from the job or environment
where the actual statement was run since it might be explained in an environment that
has different work management settings.
򐂰 If the query is implemented with multiple steps (that is, joined into a temporary file, with
grouping performed over it), the Visual Explain Only option cannot provide a valid
explanation of the SQL statement. In this case, you must use the Run and Explain
option.
8.4 Navigating Visual Explain
The Visual Explain graphics window (Figure 8-3) is presented in two parts. The left side of the
display is called the Query Implementation Graph. This is the graphical representation of the
implementation of the SQL statement and the methods used to access the database. The
arrows indicate the order of the steps. Each node of the graph has an icon that represents an
operation or values returned from an operation.
The right side of the display has the Query Attributes and Values. The display corresponds to
the object that has been selected on the graph. Initially, the query attributes and values
200
correspond to the final results icon. The vertical bar that separates the two sides is adjustable.
Each side has its own window and is scrollable.
Figure 8-3 Visual Explain Query Implementation Graph and Query Attributes and Values
The default settings cause the display to be presented with the final result icon (a checkered
flag) on the left of the display. Each icon on the display has a description and the estimated
number of rows to be used as input for each stage of the implementation.
Clicking any of the icons causes the Query Attributes and Values display to change and
presents the details that are known to the query for that part of the implementation. You might
find it helpful to adjust the display to see more of the attributes and values. Query attributes
and values are discussed further in 8.4.5, “Visual Explain query attributes and values” on
page 211.
When you right-click any of the icons on the display, an action menu is displayed. The action
menu has options to assist with query information. It can provide a shortcut to table
information to be shown in a separate window. You can find more details in 8.4.2, “Action
menu items” on page 206.
You might find the following action menu items selectively on different icons:
򐂰 Table Description: Displays table information returned by the Display File Description
(DSPFD) command
򐂰 Index Description: Displays index information returned by the DSPFD command
򐂰 Create Index: Creates a permanent index on the iSeries server
201
򐂰 Table Properties: Displays object properties
򐂰 Index Properties: Displays object properties
򐂰 Environment Settings: Displays environment settings used during the processing of this
query
򐂰 Additional fly-over panels: Exist for many of the icons
By moving the mouse pointer over the icon, a window appears with summary information
about the specific operation. See Figure 8-4.
Figure 8-4 Final Select Flyover window
The Visual Explain toolbar (Figure 8-5) helps you to navigate the displays. The first four icons
(from left to right after the printer icon) help you to control the sizing of the display. The
left-most icon scales the graphics to fit the main window. For many query implementations,
this leaves the graphical display too small to be of value. The next two icons allow you to
zoom in and out of the graphic image.
Figure 8-5 Visual Explain toolbar
202
The fourth icon (Overview) creates an additional window (Figure 8-6) that shows the Visual
Explain graphic on a reduced scale. This window has a highlighted area, which represents
the part of the image that is currently displayed in the main window.
In the Overview window (Figure 8-6), you can move the cursor into this highlighted area that
is shown in the main window. The mouse pointer changes so you can drag the highlighted
area to change the section of the overall diagram that is shown in the main window.
Figure 8-6 Visual Explain Overview window
The default schematic shows the query with the result on the left, working across the display
from right to left, to allow you to start at the result and work backward. The remaining four
icons on the Visual Explain toolbar allow you to rotate the query implementation image. The
icons are:
򐂰
򐂰
򐂰
򐂰
Starting from the right, leading to the result on the left (default view)
Starting from the left, leading to the result on the right
Starting at the bottom, leading to the result at the top
Starting from the top, leading to the result at the bottom
Try these icons to see which style of presentation you prefer. Starting in V5R1, a frame at the
bottom of the main Visual Explain window was added. In this frame, you can see two tabs.
The Statement Text tab shows the analyzed SQL statement. Also in V5R1, when Visual
Explain is used, it activates the Include Debug Messages in Job Log option and conveniently
presents those messages under the Optimizer Messages tab.
203
You can use the last icon (three steps), Statistics and Index Advisor (new in V5R2), to identify
missing or stale statistics and to specify how the statistics will be collected (Figure 8-7).
Figure 8-7 Statistics and Index Advisor: Statistics Advisor tab
Additionally, the query optimizer analyzes the row selection in the query and determines,
based on default values, if creation of a permanent index improves performance. If the
optimizer determines that a permanent index is beneficial, it returns the key columns
necessary to create the suggested index. On the Index Advisor tab, you can see the indexes
that are recommended for creation as shown in Figure 8-8. You can select the index that you
want and click the Create button to create the index selected.
Figure 8-8 Index Advisor in Visual Explain
204
8.4.1 Menu options
The menu options above the toolbar icons are File, View, Actions, Options, and Help. Starting
with V5R1, the ability to either print or save the Visual Explain output as an SQL Performance
Monitor file was added. The View options generally replicate the toolbar icons. The additional
options are:
򐂰 Icon spacing (horizontal or vertical) changes the size of the arrows between the icons.
򐂰 Arrow labels allow you to show or hide the estimated number of rows, processing time, or
the degree of parallelism that the query is processing at each stage of the implementation.
򐂰 Icon labels allow you to show or hide the description of the icons and the object name
associated with the icon.
򐂰 You can highlight expensive icons by number of returned rows and processing time.
򐂰 You can also highlight advised indexes and icons in your graph where the optimizer
recommends that you create a permanent index. The icon is highlighted in the graph and
the relevant text fields are highlighted in the Icon attributes and values table as shown in
Figure 8-9.
Figure 8-9 Highlighting Index advised information
205
8.4.2 Action menu items
The Actions menu item replicates the features that are available on the display. When you
right-click a query implementation icon, a menu appears that offers further options. These
options might include one of more of the items in the following sections.
Table Definition
The Table Definition (Table properties in V5R1) menu item opens the same window that is
used to create the table. It enables you to change the definition of the table including columns,
constraints, and partitions.
Table Description
The Table Description menu item takes you into the graphical equivalent of the DSPFD
command. From here, you can learn more information about the file. The description has
several tabs to select to find further information. A limited number of changes can be made
from the different tab windows.
Index Description
The Index Description attributes can be accessed to obtain further information about the
index. Several changes are allowed to an index from these windows, including access path
maintenance settings.
Index Definition
The Index Definition window shows the columns that exist in the table. A sequential number is
placed next to the columns that form the index, with an indication of whether the index is
ascending or descending. The display also shows the type of index.
206
Create Index
From the temporary index icon, the Create Index menu item takes you to a window where the
attributes of the temporary index are completed as shown in Figure 8-10. Simply click OK to
create a permanent index.
You need to enter an index and schema name. The type of index is assumed to be binary
radix with nonunique keys.
Note: The Create Index menu item is available from any icon where an index is advised
(for example, table scan, key positioning, key selection) in addition to the temp index icon.
This is one of the easy-to-use features of Visual Explain. It gives you the ability to easily
create an index that the optimizer has suggested.
Figure 8-10 Creating an index from Visual Explain
207
Statistic Data
Use the Statistics Data window (Figure 8-11) to view statistical data for the columns in the
selected table. The statistical information can be used by the query optimizer to determine the
best access plan for a query. The data presented here can be used to decide whether to
collect more statistics, delete existing statistics for a column, or to view more detailed column
statistics data.
Figure 8-11 Statistic Data window
208
8.4.3 Controlling the diagram level of detail
You can select how much detail you want to see on the Visual Explain graphs. The menu
options enable you to change the level of detail.
Click Options → Graph Detail → Basic to see only the icons that are directly related to the
query. Or click Options → Graph Detail → Full to see the icons that are indirectly related to
the query, such as table scans performed to build temporary indexes. If you select Full, a
white cross is displayed next to some icons where you can right-click the icon and select
Expand or Expand to window to view a subgraph of the query operation as shown in
Figure 8-12.
Figure 8-12 Controlling the diagram level of detail: *Full level (part 1 of 2)
209
Figure 8-13 shows a subgraph of the query operation. You can right-click the subgraph and
select the Collapse option to see the graph in the original way.
Figure 8-13 Controlling the diagram level of detail: *Full level (part 2 of 2)
Most users are satisfied with the *BASIC diagram, while others, with more performance
tuning experience, might prefer the *FULL diagram.
Additionally, you can view attributes in Visual Explain in two detail settings: Basic and Full (If
Possible). Select the Basic setting to retrieve only the basic subset of attributes for each icon.
Select the Full (If Possible) setting to retrieve a full set of attributes for each icon. The full set
of attributes for each icon are shown, unless there is too much data to be returned in Visual
Explain. If this happens, basic attributes are retrieved.
210
8.4.4 Displaying the query environment
The query environment is available as a fast path from the Final Results icon. It shows the
work management environment (Figure 8-14) where the query was executed. You can also
obtain this information from the Query Attributes and Values displays.
Figure 8-14 Environment
8.4.5 Visual Explain query attributes and values
The query attributes and values show more information about the optimizer implementation of
the query. If you select an icon from the Query Implementation graph, you obtain information
about that icon, as well as that part of the query implementation.
We selected a few of the query implementation icons to show you the query attributes and
values. This way you can see exactly how much information Visual Explain collects. Prior to
Visual Explain, the information was often available, but never in one place.
Table name, base table name, index name
The section in Figure 8-15 shows the name and library or schema of the table being selected.
Figure 8-15 Table name
211
Estimated processing time and table info
The estimated processing time (Figure 8-16) shows the time that the optimizer expects to
take from this part of the query. The Additional Table Info section shows information about the
rows and table size.
Figure 8-16 Estimated processing time and table information
Estimated rows selected and query join info
The estimated rows selected (Figure 8-17) shows the number of rows that the optimizer expects
to output from this part of the query. If the query is only explained, it shows an estimate of the
number of rows. If it is run and explained, it shows the number of rows that are selected.
Figure 8-17 Estimated rows selected and query join info
Estimated Cost Information About the Plan Performed
This section (Figure 8-18) indicates whether the query is CPU or I/O bound. Queries can be
CPU-intensive or I/O-intensive. When a query’s constraint resource is the CPU, it is called
CPU bound. When a query’s constraint resource is the I/O, it is called I/O bound. A query that
is either CPU or I/O bound gives you the opportunity to review the query attributes being used
when the query was processing. If a symmetric multiprocessor (SMP) is installed on a
multiprocessor system, you should review the DEGREE parameter to ensure that you are
using the systems resources effectively.
Figure 8-18 Estimated Cost Information
212
Information about the index scan performed
This display provides essential information about the index that was used for the query,
including the reason for using the index, how the index is being used, and static index
attributes. It also specifies the access method or methods used such as Index Scan - Key
positioning, Index Scan - Key Selection, and Index Only Access. To find a description about
the different reason codes, refer to the manual DB2 UDB for iSeries Database Performance
and Query Optimization for V5R3, which is available in the Information Center at:
SMP parallel information
The SMP information (Figure 8-19) shows the degree of parallelism that occurred on this
particular step. It might appear for more than one icon, because multiple steps can be
processed with differing degrees of parallelism. This information is relevant only when the
DB2 SMP licensed feature is installed.
Figure 8-19 SMP parallel information
Index advised information
The Index advised section (Figure 8-20) tells you whether the query optimizer is advising the
creation of a permanent index. If an index is being advised, the number and names of the
columns to create the index are suggested. This is the same information that is returned by
the CPI432F optimizer message. If the Highlight Index Advised option is set, advised index
information, such as base table name, library, and involved columns, are easily identifiable.
Figure 8-20 Index advised
It is possible for the query optimizer to not use the suggested index, if one is created. This
suggestion is generated if the optimizer determines that a new index might improve the
performance of the selected data by one microsecond.
213
Additional information about SQL statement
The display in Figure 8-21 shows information about the SQL environment that was used
when the statement was captured. The SQL environment parameters can impact query
performance. Many of these settings are taken from the Open Database Connectivity (ODBC)
and Java Database Connectivity (JDBC) driver settings.
Statement is Explainable specifies whether the SQL statement can be explained by the Visual
Explain tool.
Figure 8-21 Additional information
8.5 Using Visual Explain with Database Monitor data
Performance Monitor data is query information that has been recorded by one of the DB2
Universal Database for iSeries Performance Monitors into a database table that can be
analyzed later. Multiple Performance Monitors might run on the iSeries at the same time.
They can either record information for individual jobs or for the entire system. Each one is
individually named and controlled. Any given job can be monitored by a maximum of one
system monitor and one job monitor.
You can start Performance Monitors in iSeries Navigator or by using a CL command. With
iSeries Navigator, the SQL Performance Monitors component is used to collect Database
Monitor data. If you want to use Visual Explain with the data collected with an SQL
Performance Monitor, then you must choose the Detailed Monitor collection when setting up
the SLQ Performance Monitor in iSeries Navigator.
If you intend to use Visual Explain on the Database Monitor data collected with the CL
commands, the data must be imported into iSeries Navigator as detailed data.
Using Visual Explain
In iSeries Navigator, click Databases and expand the database that you want to use. Click
SQL Performance Monitors to obtain a list of the SQL Performance Monitors that are
currently on the system.
Right-click the name that you want, and select List Explainable Statements. An explainable
statement (Figure 8-22) is an SQL statement that can be explained by Visual Explain.
Because Visual Explain does not process all SQL statements, some statements might not be
selected.
214
Figure 8-22 SQL explainable statements
The explainable SQL statements that were optimized by the job are now listed. If you were
monitoring an SQL Script window, this is the SQL statement that was entered.
Note: Query optimizer information is generated only for an SQL statement or query
request when an open data path (ODP) is created. When an SQL or query request is
implemented with a reusable ODP, then the query optimizer is not invoked. Therefore,
there is no feedback from the query optimizer in terms of monitor data or debug messages.
Also, the statement is not explainable in Visual Explain. The only technique for analyzing
the implementation of a statement in reusable ODP mode is to look for an earlier execution
of that statement when an ODP was created for that statement.
To use Visual Explain on any of the statements, select the statement from the display. The full
SQL statement appears in the lower part of the display for verification. Click Run Visual
Explain (Figure 8-22) to analyze the statement and prepare a graphical representation of the
query.
Exit the Visual Explain window and the Explainable Statements window when you have
completed your analysis. You might either retain the performance data or remove it from the
system at this time, depending on your requirements.
215
8.6 Using Visual Explain with imported data
You can import Database Monitor data into Visual Explain and then use the tool to help with
diagnosing problems further. Visual Explain can be used against current active jobs and
against data collected in Performance Monitors either by iSeries Navigator or using the
STRDBMON command.
To import a Database Monitor from another system or the same system, you can use iSeries
Navigator. Select the system where the data is held. Click Databases → your relational
database. Right-click SQL Performance Monitors and select Import as shown in
Figure 8-23.
Figure 8-23 Selecting to import a Database Monitor
In the Import SQL Performance Monitor Files window (Figure 8-24), specify the name of the
monitor as it will be known in iSeries Navigator and specify the file name and schema in
which it resides. For Type of monitor, select either Summary or Detailed depending on how
you collected the Database Monitor data.
Figure 8-24 Import SQL Performance Monitor Files window
216
After the monitor is imported, it is displayed in the right pane of the iSeries Navigator window
as shown in Figure 8-25.
Figure 8-25 SQL Performance Monitors window
After you import the monitor, you can either choose either List Explainable Statements or
Analyze results.
217
8.6.1 List Explainable Statements
When you see the SQL Performance Monitor of interest in the iSeries Navigator window,
right-click and select List Explainable Statements. A window opens that shows the list of
explainable statements (see Figure 8-26). You can select an SQL statement by clicking the
required row. You can sort on the columns to help you look for statements of interest, like long
running statements as shown in Figure 8-26.
Figure 8-26 List Explainable Statements sorted by most expensive
218
The SQL statement appears in the bottom half of the window. To obtain the visual
explanation, click the Run Visual Explain button. After you select an SQL statement for
analysis, you can use Visual Explain to understand the implementation the optimizer chose,
the time it took to run, any indexes advised, the environmental variables that were used, and
the resources that were available at the time the job ran as shown in Figure 8-27.
Figure 8-27 Final Select Visual Explain window
8.7 Non-SQL interface considerations
Obviously, the SQL Performance Monitor can capture implementation information for any
SQL-based interface. Therefore, any SQL-based request can be analyzed with Visual
Explain. SQL-based interfaces range from embedded SQL to Query Manager reports to
ODBC and JDBC.
Some query interfaces on the iSeries servers are not SQL-based and, therefore, are not
supported by Visual Explain. The interfaces not supported by Visual Explain include:
򐂰
򐂰
򐂰
򐂰
Native database access from a high level language, such as Cobol, RPG, and so on
Query
OPNQRYF command
OS/400 Create Query API (QQQQRY)
The query optimizer creates an access plan for all queries that run on the iSeries server. Most
queries use the SQL interface. They generate an SQL statement, either directly (SQL Script
219
window, STRSQL command, SQL in high-level language (HLL) programs) or indirectly (Query
Monitor/400).
Other queries do not generate identifiable SQL statements (Query, OPNQRYF command)
and cannot be used with Visual Explain via the SQL Performance Monitor. In this instance,
the name SQL, as part of the SQL Performance Monitor, is significant.
The statements that generate SQL and that can be used with the Visual Explain via the SQL
Performance Monitor include:
򐂰
򐂰
򐂰
򐂰
򐂰
SQL statements from the SQL Script Center
SQL statements from the Start SQL (STRSQL) command
SQL statements processed by the Run SQL Statement (RUNSQLSTM) command
SQL statements embedded into a high level language program (Cobol, Java, or RPG)
SQL statements processed through an ODBC or JDBC interface
The statements that do not generate SQL and, therefore, that cannot be used with Visual
Explain via the SQL Performance Monitor include:
򐂰
򐂰
򐂰
򐂰
Native database access from a high level language, for example, Cobol, RPG, and so on
Query
Open Query File (OPNQRYF) command
OS/400 Create Query API (QQQQRY)
8.8 The Visual Explain icons
Table 8-1 lists the icons that you might find on the Visual Explain query implementation chart.
Table 8-1 Visual Explain icons
The Final Select icon displays the original text and summary information about how the query
was implemented.
The Table icon indicates that a table was accessed. See the Select icon for more details.
The Table Scan icon indicates that all rows in the table were paged in, and selection criteria
was applied against each row. Only those rows that meet the selection criteria were
retrieved. To obtain the result in a particular sequence, you must specify the ORDER BY
clause.
The Table Probe icon indicates that data from the table must be processed and selected for
this query. The table is probed using a key derived from the ordinal number or relative record
number associated with each row in the table. The ordinal number is used to calculate the
pages of data that need to be retrieved and brought into main memory to satisfy this probe
request. The ordinal number used for the probe operation was provided by another data
access method previously processed for this query.
The Index icon indicates that an index object was used to process this query.
The Index Scan icon indicates that the entire index will be scanned, which causes all of the
entries in the index that are to be paged into main memory to be processed. Any selection
criteria whose predicates match the key columns of the index can then be applied against
the index entries. Only those key entries that match the specified key selection are used to
select rows from the corresponding table data.
220
The Index Probe icon indicates that the selection criteria whose predicates matched the
leading key columns of the index were used to probe directly into the index. The probe
selection minimizes the number of key entries that must be processed and paged into main
memory. Additional key selection can be applied against the non-leading key columns of the
index to further reduce the number of selected key entries. Only key entries that match the
specified probe and key selection are used to select rows from the corresponding table data.
The Parallel Table Scan icon indicates that a table scan access method was used and multiple
tasks were used to fill the rows in parallel. The table was partitioned, and each task was
given a portion of the table to use.
The Skip Sequential Table Scan icon indicates that a bitmap was used to determine which rows
were selected. No CPU processing was done on non-selected rows, and I/O was minimized
by bringing in only those pages that contained rows to be selected. This icon is usually
related to the Dynamic Bitmap or Bitmap Merge icons.
The Skip Sequential Parallel Scan icon indicates that a skip sequential table scan access
method was used and multiple tasks were used to fill the rows in parallel. The table was
partitioned, and each task was given a portion of the table to use.
The Derived Column Selection icon indicates that a column in the row selected had to be
mapped or derived before selection criteria could be applied against the row. Derived
column selection is the slowest selection method.
The Parallel Derived Column Selection icon indicates that derived field selection was
performed, and the processing was accomplished using multiple tasks. The table was
partitioned, and each task was given a portion of the table to use.
The Index Key Positioning icon indicates that only entries of the index that match a specified
range of key values were “paged in”. The range of key values was determined by the
selection criteria whose predicates matched the key columns of the index. Only selected key
entries were used to select rows from the corresponding table data.
The Parallel Index Key Positioning icon indicates that multiple tasks were used to perform the
key positioning in parallel. The range of key values was determined by the selection criteria,
whose predicates matched the key columns of the index. Only selected key entries were
used to select rows from the corresponding table data.
The Index Key Selection icon indicates that all entries of the index were paged in. Any
selection criteria whose predicates match the key columns of the index was applied against
the index entries. Only selected key entries were used to select rows from the table data.
The Parallel Index Key Selection icon indicates that multiple tasks were used to perform key
selection in parallel. The table was partitioned, and each task was given a portion of the table
to use.
The Encoded-vector Index icon indicates that access was provided to a database file by
assigning codes to distinct key values, and then representing these values in an array
(vector). Because of their compact size and relative simplicity, encoded-vector indexes
(EVIs) provide for faster scans.
The Parallel Encoded-vector Index icon indicates that multiple tasks were used to perform the
EVI selection in parallel. This allows for faster scans that can be more easily processed in
parallel.
The Encoded-vector Index Scan icon indicates that the entire EVI will be scanned causing all
of the distinct values represented in the index to be processed. Any selection criteria, whose
predicates match the key columns of the EVI can then be applied against the distinct values
represented in the index. Only those distinct values that match the specified key selection
are then used to process the vector and generate either a temporary row number list or
temporary row number bitmap.
221
The Encoded-vector Index Probe icon indicates that the selection criteria whose predicates
matched the leading key columns of the EVI were used to probe directly into the distinct
values represented in the index. Only those distinct values that match the specified probe
selection are then used to process the vector and generate either a temporary row number
list or temporary row number bitmap.
The Sort Sequence icon indicates that selected rows were sorted using a sort algorithm.
The Grouping icon indicates that selected rows were grouped or summarized. Therefore,
duplicate rows within a group were eliminated.
The Nested Loop Join icon indicates that queried tables were joined together using a nested
loop join implementation. Values from the primary file were joined to the secondary file by
using an index whose key columns matched the specified join columns. This icon is usually
after the method icons used on the underlying tables (that is, index scan-key selection and
index scan-key positioning).
The Hash Join icon indicates that a temporary hash table was created. The tables queried
were joined together using a hash join implementation where a hash table was created for
each secondary table. Therefore, matching values were hashed to the same hash table
entry.
The Temporary Index icon indicates that a temporary index was created, because the query
either requires an index and one does not exist, or the creation of an index will improve
performance of the query.
The Temporary Hash Table icon indicates that a temporary hash table was created to perform
hash processing.
The Temporary Table icon indicates that a temporary table was required to either contain the
intermediate results of the query, or the queried table could not be queried as it currently
exists and a temporary table was created to replace it.
The Dynamic Bitmap icon indicates that a bitmap was dynamically generated from an existing
index. It was then used to determine which rows were to be retrieved from the table. To
improve performance, dynamic bitmaps can be used in conjunction with a table scan access
method for skip sequential or with either the index key position or key selection.
The Bitmap Merge icon indicates that multiple bitmaps were merged or combined to form a
final bitmap. The merging of the bitmaps simulates boolean logic (AND/OR selection).
The DISTINCT icon indicates that duplicate rows in the result were prevented. You can
specify that you do not want any duplicates by using the DISTINCT keyword, followed by the
selected column names.
The UNION Merge icon indicates that the results of multiple subselects were merged or
combined into a single result.
The Subquery Merge icon indicates that the nested SELECT was processed for each row
(WHERE clause) or group of rows (HAVING clause) selected in the outer level SELECT. This
is also referred to as a correlated subquery.
The Hash Table Scan icon indicates that the entire temporary hash table will be scanned and
all of the entries contained with the hash table will be processed. A hash table scan is
generally considered when optimizer is considering a plan that requires the data values to
be collated together but the sequence of the data is not required. The use of a hash table
scan allows the optimizer to generate a plan that can take advantage of any non-join
selection while creating the temporary hash table.
222
The Hash Table Probe icon indicates that the selection criteria that match the key columns
used to create the temporary hash table will be probed to find all of the matching values
stored within the hash table. A hash table probe is generally considered when determining
the implementation for a secondary table of a join. The use of a hash table probe allows the
optimizer to generate a plan that can take advantage of any non-join selection while creating
the temporary hash table. An additional benefit of using a hash table probe is that the data
structure of the temporary hash table usually causes the table data to remain resident within
main memory after creation, reducing paging on the subsequent probe operation.
The Temporary Distinct Hash Table icon indicates that a temporary distinct hash table was
created in order to perform hash processing. A distinct hash table is a data structure that is
identical to the temporary hash table, except all duplicate data is compressed out of the
temporary being created. The resulting hash table can then be used to perform distinct or
aggregate operations for the query.
The Distinct Hash Table Scan icon indicates that the entire temporary distinct hash table will
be scanned and all of the entries contained with the hash table will be processed. A distinct
hash table scan is generally considered when optimizer is considering a plan that requires
the data values to be collated together and all duplicate removed but the sequence of the
data is not required. The use of a distinct hash table scan allows the optimizer to generate
a plan that can take advantage of any non-join selection while creating the temporary distinct
hash table. An additional benefit of using a distinct hash table scan is that the data structure
of the temporary distinct hash table usually causes the table data within the distinct hash
table to remain resident within main memory after creation. This benefit reduces the paging
on the subsequent scan operations.
The Distinct Hash Table Probe icon indicates that the selection criteria that match the key
columns used to create the temporary distinct hash table will be probed to find all of the
matching values stored within the hash table. The use of a distinct hash table probe allows
the optimizer to generate a plan that can take advantage of any non-join selection while
creating the temporary distinct hash table. An additional benefit of using a distinct hash table
probe is that the data structure of the temporary distinct hash table usually causes the table
data to remain resident within main memory after creation. This benefit reduces the paging
on the subsequent probe operation.
The Temporary Sorted List icon indicates that a temporary sorted list was created in order to
perform a sequencing operation. A sorted list is a data structure where the table data is
collated and sorted based upon the value of a column or columns referred to as the sort key.
The sorted list can then be used to return the data in a specified sequence or to perform
probe operations using the sort key to quickly retrieve all of the table data that matches a
particular sort key.
The Sorted List Scan icon indicates that the entire temporary sorted list will be scanned and
all of the entries contained with the sorted list will be processed. A sorted list scan is
generally considered when optimizer is considering a plan that requires the data values to
be sequenced based upon the sort key of the sorted list. The use of a sorted list scan allows
the optimizer to generate a plan that can take advantage of any non-join selection while
creating the temporary sorted list. An additional benefit of using a sorted list scan is that the
data structure of the temporary sorted list usually causes the table data within the sorted list
to remain resident within main memory after creation. This benefit reduces the paging on
the subsequent scan operations.
223
The Sorted List Probe icon indicates that the selection criteria that match the key columns
used to create the temporary sorted list is probed to find all of the matching values stored
within the sorted list. A sorted list probe is generally considered when determining the
implementation for a secondary table of a join when either the join condition uses an
operator other than equal or a temporary hash table is not allowed in this query environment.
The use of a sorted list probe allows the optimizer to generate a plan that can take
advantage of any non-join selection while creating the temporary sorted list. An additional
benefit of using a sorted list probe is that the data structure of the temporary sorted list
usually causes the table data to remain resident within main memory after creation. This
benefit reduces the paging on the subsequent probe operation.
The Temporary List icon indicates that a temporary list was created. The temporary list was
required to either contain the intermediate results of the query, or the queried table could not
be queried as it currently exists and a temporary list was created to replace it. The list is an
unsorted data structure with no key. The data contained within the list can only be retrieved
by a scan operation.
The List Scan icon indicates that the entire temporary list will be scanned and all of the entries
will be processed.
The Temporary Row Number List icon indicates that a temporary row number list was created
in order to help process any selection criteria. A row number list is a data structure used to
represent the selected rows from a table that matches any specified selection criteria. Since
the selected rows are represented by a sorted list of row numbers, multiple lists can be
merged and combined to allow for complex selection and processing to be performed
without having any paging occur against the table itself.
The Row Number List Scan icon indicates that the entire row number list will be scanned and
all of the entries will be processed. Scanning a row number list can provide large amounts
of savings for the table data associated with the temporary row number list. Since the data
structure of the temporary row number list guarantees that the row numbers are sorted, it
closely mirrors the row number layout of the table data, ensuring that the paging on the table
will never revisit the same page of data twice.
The Row Number List Probe icon indicates that a row number list was used to verify that a row
from a previous operation in the query matches the selection criteria used to create the
temporary row number list. The use of a row number list probe allows the optimizer to
generate a plan that can process the rows in the table in any manner regardless of any
specified selection criteria. As the rows are processed, the ordinal number from the row is
used to probe into the row number list to determine if that row matches the selection criteria.
This is generally found when an index is used to satisfy the ORDER BY from a query and a
separate viable index exists to process the selection criteria.
The Bitmap Scan icon indicates that the entire bitmap will be scanned and all of the entries
that represent selected rows will be processed. Scanning a bitmap can provide large
amounts of savings for the table data associated with the temporary bitmap. Since the data
structure of the temporary bitmap mirrors the row number layout of the table data, the bitmap
can be used to efficiently schedule paging of the table for all selected rows.
The Bitmap Probe icon indicates that a bitmap was used to verify that a row from a previous
operation in the query matches the selection criteria used to create the temporary bitmap.
The use of a bitmap probe allows the optimizer to generate a plan that can process the rows
in the table in any manner regardless of any specified selection criteria. As the rows are
processed, the ordinal number from the row is used to probe into the bitmap to determine if
that row matches the selection criteria. This is generally found when an index is used to
satisfy the ORDER BY from a query and a separate viable index exists to process the
selection criteria.
224
The Index Scan icon indicates that the entire temporary index will be scanned causing all of
the entries in the index to be paged into main memory to be processed. Any selection criteria
whose predicates match the key columns of the index can then be applied against the index
entries. Only those key entries that match the specified key selection are used to select rows
from the corresponding table data.
The Index Probe icon indicates that the selection criteria whose predicates matched the
leading key columns of the index were used to probe directly into the temporary index. The
probe selection minimizes the number of key entries that must be processed and paged into
main memory. Additional key selection can be applied against the non-leading key columns
of the temporary index to further reduce the number of selected key entries. Only those key
entries that matched the specified probe and key selection are used to select rows from the
corresponding table data.
The Temporary Correlated Hash Table icon indicates that a temporary correlated hash table
was created in order to perform hash processing. A hash table is a data structure where the
table data is collated based upon the value of a column or columns referred to as the hash
key. The hash table can then be used to perform probe operation using the hash key to
quickly retrieve all of the table data that matches a particular hash value. Because this is a
correlated hash table, the hash table needs to be rebuilt prior to any scan or probe
operations being performed.
The Correlated Hash Table Scan icon indicates that the entire temporary hash table will be
scanned and all of the entries contained with the hash table will be processed. A correlated
hash table scan is generally considered when optimizer is considering a plan that requires
the data values to be collated together but the sequence of the data is not required. In
addition, the some of the values used to create the correlated hash table can change from
one scan to another. The use of a correlated hash table scan allows the optimizer to
generate a plan that can take advantage of any non-join selection while creating the
temporary correlated hash table. An additional benefit of using a correlated hash table scan
is that the data structure of the temporary correlated hash table usually causes the table
data within the hash table to remain resident within main memory after creation. This benefit
reduces the paging on the subsequent scan operations.
The Correlated Hash Table Probe icon indicates that the selection criteria that match the key
columns used to create the temporary correlated hash table will be probed to find all of the
matching values stored within the hash table. A correlated hash table probe is generally
considered when determining the implementation for a secondary table of a join. The use of
a hash table probe allows the optimizer to generate a plan that can take advantage of any
non-join selection while creating the temporary correlated hash table. An additional benefit
of using a correlated hash table probe is that the data structure of the temporary correlated
hash table usually causes the table data to remain resident within main memory after
creation. This benefit reduces paging on the subsequent probe operation.
The Temporary Correlated List icon indicates that a temporary correlated list was created. The
temporary correlated list was required to either contain the intermediate results of the query,
or the queried table could not be queried as it currently exists and a temporary correlated
list was created to replace it. The list is an unsorted data structure with no key that must be
rebuilt prior to any scan operation being performed.
The Correlated List Scan icon indicates that the entire temporary list will be scanned and all
of the entries will be processed.
The Temporary Buffer icon indicates that a temporary buffer was created to store the
intermediate rows of an operation. The temporary buffer is generally considered at a
serialization point within a query to help facilitate operations such as parallelism. The buffer
is an unsorted data structure, but it differs from other temporary data structures in that the
buffer does not have to be fully populated in order allow its results to be processed.
225
The Buffer Scan icon indicates that the entire temporary buffer will be scanned and all of the
entries will be processed.
The Table Random Pre-Fetch icon indicates that the pages required for the table probe
operation will be requested synchronously in the background prior to the actual table probe
operation being performed. The system attempts to manage the paging for the table probe
to maintain that all of the pages of data necessary to perform the table probe operation stay
resident within main memory until they are processed. The amount of pre-fetch paging that
is performed by this data access method is dynamically controlled by the system based upon
memory consumption and the rate at which rows continue to be processed.
The Table Clustered Pre-Fetch icon indicates that the pages required for the table probe
operation will be requested synchronously in the background prior to the actual table probe
operation being performed. The system attempts to manage the paging for the table probe
to maintain that all of the pages of data necessary to perform the table probe operation stay
The Index Random Pre-Fetch icon indicates that the pages required for the index probe
operation will be requested synchronously in the background prior to the actual index probe
operation being performed. The system attempts to manage the paging for the index probe
to maintain that all of the pages of data necessary to perform the index probe operation stay
The Logic icon indicates that the query needed to perform an operation or test against the
data in order to generate the selected rows.
The Fetch N Rows icon indicates that a limit was placed upon the number of selected rows.
The fetch n rows access method can either be used to implement a user specified limit on
the selected rows or it can be combined with other access methods by the optimizer to
satisfy complex implementation plans.
The Lock Row for Update icon indicates that an update lock was acquired for the associated
table data in order to perform an update or delete operation. To minimize contention between
queries, the optimizer attempts to place the lock row for update operation such that the lock
is not acquired and held for a long duration.
The User-defined table function icon indicates that a user-defined function that returns a table
was used. A table function can be referenced in an SQL FROM clause in the same way that
a table or view can be referenced.
The Select icon indicates a point in the query where multiple results are brought together into
a single result set. For example, if a query is the union of two different select statements, at
the point before the union occurs, the Select icon indicates the points where the select
statements finished and the union is about to occur. This icon also represents the default
information for an operation that is unknown or not defined elsewhere with Visual Explain. It
can help to represent tables or insert, update and delete operations for a query. The
summary information for this icon contains any available information to help describe the
operation being performed and what the icon represents.
The Incomplete Information icon indicates that a query could not be displayed due to
incomplete information.
226
Part 3
Part
3
Additional tips
In this part, we provide additional tips to help prevent database performance problems. We
present tips regarding indexing strategy and optimizing your SQL statements.
This part encompasses Chapter 9, “Tips to prevent further database performance problems”
on page 229.
227
228
9
Chapter 9.
Tips to prevent further database
performance problems
It is important to analyze database performance problems and to fix them. For programmers,
it is much more important to know how to prevent those problems from the beginning and how
to achieve the best performance.
In this chapter, we help you to understand how the optimizer decides for the optimal index, so
that you can predict which indexes are necessary. We also provide further tips and
information about:
򐂰 Optimizing SQL statements
򐂰 Coding techniques to avoid with SQL
򐂰 Enhancements to the Reorganize Physical File Member (RGZPFM) command
229
9.1 Indexing strategy
The database has two types of permanent objects, tables and indexes. Tables and indexes
include information about the object’s structure, size, and attributes. In addition, tables and
indexes contain statistical information about the number of distinct values in a column and the
distribution of those values in the table. The DB2 Universal Database for iSeries optimizer
uses this information to determine how to best access the requested data for a given query
request.
Since the iSeries optimizer uses cost-based optimization, the more information given about
the rows and columns in the database, the optimizer is better able to create the best possible
(least costly and fastest) access plan for the query. With the information from the indexes, the
optimizer can make better choices about how to process the request (local selection, joins,
grouping, and ordering).
The primary goal of the optimizer is to choose an implementation that quickly and efficiently
eliminates the rows that are not interesting or required to satisfy the request. Normally, query
optimization is concerned with trying to find the rows of interest. A proper indexing strategy
assists the optimizer and database engine with this task.
Note: Although you cannot specify indexes in an SQL statement, the optimizer uses them
to implement the best access plan to get access to the requested data. You can only
specify tables and views in SQL statements.
9.1.1 Access methods
As it stated previously, the database has two types of permanent objects (tables and indexes).
There are several methods or algorithms that can be used with this objects:
򐂰 Table scan
With a tables scan, all rows of a tables are processed regardless of the selectivity of the
query. Deleted records are examined even though none are selected.
򐂰 Table probe
A table probe operation is used to retrieve a specific row from a table based upon its row
number. The row number is provided to the table probe access method by some other
operation that generates a row number for the table. This can include index operations as
well as temporary row number lists or bitmaps.
򐂰 Index scan
With an index scan, all keys from the index are processed. The resulting rows are
sequenced based upon the key columns. This characteristic is used to satisfy a portion of
the query request (such as ordering or grouping).
򐂰 Index probe
An index probe reads like an index scan in a keyed sequence, but the requested rows are
first identified by a probe operation.
9.1.2 Guidelines for a perfect index
Typically you create an index for the most selective columns and create statistics for the least
selective columns in a query. By creating an index, the optimizer knows that the column is
selective, which gives the optimizer the ability to choose that index to implement the query.
230
In a perfect radix index, the order of the columns is important. In fact, it can make a difference
as to whether the optimizer uses it for data retrieval at all. As a general rule, order the
columns in an index in the following way:
򐂰 Place equal predicates first. Predicates using the equal (=) operator generally eliminate
the largest number of nonparticipating rows and should therefore be first in the index.
򐂰 If all of the predicates have an equal operator, then order the columns as follows:
a.
b.
c.
d.
Selection predicates + join predicates
Join predicates + selection predicates
Selection predicates + group by columns
Selection predicates + order by columns
򐂰 Always place the most selective columns as the first key in the index.
Note: Indexes consume system resources, so it up to you to find a balance between query
performance and system (index) maintenance.
The query shown in Example 9-1 uses the ITEMS table and finds all the customers who
returned orders at year end 2000 that were shipped via air. We illustrate the perfect indexes
for this query.
Example 9-1 One table query
SELECT CUSTOMER, CUSTOMER_NUMBER, ITEM_NUMBER
FROM ITEMS
WHERE
"YEAR"
= 2000
AND "QUARTER" =
4
AND RETURNFLAG = 'R'
AND SHIPMODE
= 'AIR'
ORDER BY CUSTOMER_NUMBER, ITEM_NUMBER
The query has four local selection predicates and two ORDER BY columns.
Following the guidelines, the perfect index places the key columns first that cover the equal
predicates (“YEAR”, “QUARTER”, RETURNFLAG, SHIPMODE), followed by the ORDER BY
columns CUSTOMER_NUMBER, ITEM_NUMBER as specified in Example 9-2.
Example 9-2 Perfect index for the one table query example
CREATE INDEX MySchema/ItemIdx01
ON MySchema/Items
("YEAR", "QUARTER", ReturnFlag, ShipMode,
Customer_Number,
Item_Number)
Encoded-vector index guidelines
An encoded-vector index (EVI) cannot be used for grouping or ordering and has a limited use
in joins. Single key EVIs can be used to create bitmaps or relative record number (RRN) lists
that can be used in combination with binary radix tree indexes. You might use this technique
when the local selection contains AND or OR conditions and a single index does not contain
all the proper key columns or a single index cannot meet all of the conditions.
If you look at the query in Example 9-1, you see four local selection predicates and two
ORDER BY columns. Following the EVI guidelines, single key indexes are created with key
columns covering the equal predicates as shown in Example 9-3.
Chapter 9. Tips to prevent further database performance problems
231
Example 9-3 EVIs for the one table query example
CREATE ENCODED VECTOR INDEX MySchema/ItemsEVI_Year
ON MySchema/Items ("YEAR");
CREATE ENCODED VECTOR INDEX MySchema/ItemsEVI_Quarter
ON MySchema/Items ("QUARTER");
CREATE ENCODED VECTOR INDEX MySchema/ItemsEVI_ReturnFlag
ON MySchema/Items (RETURNFLAG);
CREATE ENCODED VECTOR INDEX MySchema/ItemsEVI_ShipMode
ON MySchema/Items (SHIPMODE);
9.1.3 Additional indexing tips
Keep in mind that indexes are used by the optimizer for the optimization phase or the
implementation phase of the query. For example, you might create an index that you do not
see is being used by the optimizer for the implementation phase, but it might have been used
for the optimization phase. There are some additional considerations to remember where the
optimizer might not use an index:
򐂰 Avoid NULL capable columns if expecting to use index only access. When any key in the
index is NULL capable, the index-only access method cannot be used with the Classic
Query Engine (CQE).
򐂰 Avoid derived expressions in local selection. Access via an index might not be used for
predicates that have derived values. Or, a temporary index is created to provide key values
and attributes that match the derivative. For example, if a query includes one of the
following predicates, the optimizer considers that predicate to be a derived value and
might not use an index for local selection:
T1.ShipDate > (CURRENT DATE – 10 DAYS)
UPPER(T1.CustomerName) = "SMITH"
򐂰 Index access is not used for predicates where both operands come from the same table.
For example, if the WHERE clause contains the following snippet, the optimizer does not
use an index to retrieve the data since it must access the same row for both operands.
T1.ShipDate > T1.OrderDate
򐂰 Consider index only access (IOA). If all of the columns used in the query are represented
in the index as key columns, the optimizer can request index only access. With IOA, DB2
Universal Database for iSeries does not have to retrieve any data from the actual table. All
of the information required to implement the query is available in the index. This might
eliminate the random access to the table and drastically improve query performance.
򐂰 Use the most selective columns as keys in the index. Give preference to columns used in
equal comparisons.
򐂰 For key columns that are unique, use a unique constraint.
򐂰 Make sure that statistics exist for the most and least selective columns for the query.
232
9.1.4 Index Advisor
The optimizer has several feedback mechanisms to help you identify the need of an index. To
help you to identify the perfect index for a given query request, you can use the following
methods to find the requested index:
򐂰 STRDBG (Start Debug) CL command
򐂰 Running and analyzing an SQL statement using Visual Explain Index Advisor in iSeries
Navigator
STRDBG CL command
You can use the STRDBG CL command to analyze the feedback of the optimizer on the
green-screen interface. Before you start SQL by using the Start SQL (STRSQL) CL command
or before you call your program containing embedded SQL statement, you must enter the
STRDBG command. With this command, all SQL messages are registered in the job log.
If an access path, either SQL index or keyed logical file is found, you see the appropriate
message in the job log (see Figure 9-1).
Query options retrieved file QAQQINI in library QUSRSYS.
All access paths were considered for file ORDHDR.
Access path of file ORDHDRI02 was used by query.
Query options used to build the OS/400 query access plan.
Figure 9-1 Job log for an executed SQL statement where an access path is available
If no access path is available, you might find an access path suggestion. Figure 9-2 shows the
job log for an SQL statement, where an access path is suggested.
Query options retrieved file QAQQINI in library QUSRSYS.
Arrival sequence access was used for file ORDHDR.
Access path suggestion for file ORDHDR.
Query options used to build the OS/400 query access plan.
Figure 9-2 Job log for an executed SQL statement, where no access path is available
233
If you look at the detailed message for the access path suggestion (position the cursor on the
message and press F1), you see the recommended key fields as in Figure 9-3.
Additional Message Information
Message ID . . . . . . :
Message type . . . . . :
Date sent . . . . . . :
CPI432F
Information
03/15/05
Severity . . . . . . . :
00
Time sent
11:02:37
. . . . . . :
Message . . . . :
Access path suggestion for file ORDHDR.
Cause . . . . . :
To improve performance the query optimizer is suggesting a
permanent access path be built with the key fields it is recommending. The
access path will access records from member ORDHDR of file ORDHDR in library
MYSCHEMA.
In the list of key fields that follow, the query optimizer is recommending
the first 1 key fields as primary key fields. The remaining key fields are
considered secondary key fields and are listed in order of expected
selectivity based on this query. Primary key fields are fields that
significantly reduce the number of keys selected based on the corresponding
selection predicate. Secondary key fields are fields that may or may not
significantly reduce the number of keys selected. It is up to the user to
determine the true selectivity of secondary key fields and to determine
whether those key fields should be used when creating the access path.
The query optimizer is able to perform key positioning over any
combination of the primary key fields, plus one additional secondary key
field. Therefore it is important that the first secondary key field be the
most selective secondary key field. The query optimizer will use key
selection with any remaining secondary key fields. While key selection is
not as fast as key positioning it can still reduce the number of keys
selected. Hence, secondary key fields that are fairly selective should be
included. When building the access path all primary key fields should be
specified first followed by the secondary key fields which are prioritized
by selectivity. The following list contains the suggested primary and
secondary key fields:
ORDER_DATE.
If file ORDHDR in library MYSCHEMA is a logical file then the access
path should be built over member ORDHDR of physical file ORDHDR in library
MYSCHEMA.
Figure 9-3 Access path suggestion: Detailed information
234
Running and analyzing an SQL statement using Visual Explain Index
Advisor in iSeries Navigator
If you execute an SQL script or use Performance Monitors, you can use Visual Explain to
analyze your SQL statements. Visual Explain has a feature, called Index Advisor, that might
recommend the required indexes and offers a way to create them easily.
To execute and analyze an SQL statement:
1. Start iSeries Navigator.
2. Under My Connections, click your connection → Databases.
3. Select your database and, in the Database tasks area in the bottom right pane, select
Run an SQL Script. See Figure 9-4.
Figure 9-4 Selecting Run an SQL script
235
4. In the Run an SQL script window, type your SQL statement and highlight it. From the
toolbar, click the Visual Explain option and select either Explain or Run Explain.
Figure 9-5 illustrates these steps to start Visual Explain for a specified SQL statement.
Figure 9-5 Start Visual Explain
236
5. The window in Figure 9-6 shows the Visual Explain presentation of the SQL statement.
Note the joblog messages in the bottom pane of the window. In this example, you see that
a table scan is performed, but an index is suggested. To see which index is recommended,
click Actions and select Advisor.
Figure 9-6 Starting the Index Advisor
237
The Index and Statistics Advisor window (Figure 9-7) shows you all the recommended
indexes and statistic columns. If you click Create, you can easily create the suggested
indexes. All key columns are listed in the desired sequence.
Figure 9-7 Index Advisor showing the recommended indexes
9.2 Optimization of your SQL statements
It is not possible to force the optimizer to use a particular index, but you can affect the
optimizer’s decision by coding the SQL select statement in different ways. In the following
section, we show you some ways that you can affect the optimizer.
9.2.1 Avoid using logical files in your select statements
It is a common misunderstanding that, by specifying logical files in SQL statements, the query
optimizer can be forced to choose this index. This is not true. The specified index can be
selected if it meets all the requirements for the best access path. But the query optimizer can
choose any other index or decide even on doing a table scan as well.
If you specify a keyed logical file in an SQL select statement, the optimizer takes only the
column selection, join information, and select/omit clauses, and rewrites the query. For each
of these joined tables, a specific index is determined.
There is another point to avoid specifying logical files in an SQL statement. A select
statement that contains logical files is rerouted and executed by the CQE and does not
benefit from the enhancements of the new SQL Query Engine (SQE). The cost of rerouting
might cause an overhead of up to 10 to 15% in the query optimization time.
238
We illustrate this with an example. Suppose that we have a base table ORDHDR that contains
all order header information. In addition to a primary key constraint, two logical files are
created.
Example 9-4 show the DDS description of the keyed logical file ORDHDRL1, with the key
fields ORHDTE = Order Date and ORHNBR = Order Number.
Example 9-4 Logical file ORDHDRL1
A
R ORDHDRF
PFILE(ORDHDR)
*
A
A
K ORHDTE
K ORHNBR
Example 9-5 shows the DDS description of the second logical file ORDHDRL2, with the key
fields ORHDLY = Delivery Date in descending order, CUSNBR = Customer Number, and
ORHNBR = Order Number.
Example 9-5 Logical file ORDHDRL2
A
R ORDHDRF
PFILE(ORDHDR)
K ORHDLY
K CUSNBR
K ORHNBR
DESCEND
*
A
A
A
Now we want to select all orders with the order date 11 March 2005. In the first case, we
perform the select using the logical file ORDHDRL2, and in the second case, we use the
physical file.
239
Figure 9-8 shows the Visual Explain for both statements. In both cases, the access path of
the keyed logical file is used. But in the first case, an Index Scan - Key Positioning is
performed that refers to the CQE, while in the second case, an index probe in combination
with a table probe is executed that refers to SQE.
Figure 9-8 Comparing the use of logical and physical files in select statements
9.2.2 Avoid using SELECT * in your select statements
When reading a record/row with native I/O, all fields/columns and field/column values are
always moved into memory. If you have tables with a lot of columns and you only need
information from a few, a lot of unnecessary information must be loaded. With SQL, you can
select only the columns that you need to satisfy the data request.
If you specify the required columns in your select statements, the optimizer might perform
index only access. With IOA, DB2 Universal Database for iSeries does not have to retrieve
any data from the actual table. All of the information required to implement the query is
available in the index. This might eliminate the random access to the table and drastically
improve query performance.
240
9.2.3 Avoid using the relative record number to access your data
For most tables, it is easy to define a primary or unique key. In some cases, such as for
transaction tables, it is almost impossible to determine a unique key. With native I/O, the
required records are often read using the relative record number. For native I/O, this is the
fastest access method.
In SQL, the relative record number can be determined by using the scalar function RRN().
The relative record number is not a defined row in the database table. Therefore, it is not
possible to create an index over the relative record number. If you try to read a row using the
relative record number, a table scan is always performed.
For tables where you cannot specify a unique key, you must create an additional column to
hold a unique value. SQL provides methods, such as identity columns or columns with
ROWID data type, where a unique value is automatically generated if a new row is inserted.
Identity columns or columns with ROWID data type are ideal for primary or unique key
constraints. However, it is also possible to create an index over these columns.
Figure 9-9 shows the selection of a specific row using the SQL scalar function RRN().
Figure 9-9 Row selection by using the SQL scalar function RRN()
241
9.2.4 Avoid numeric data type conversion
When a column value and a host variable (or constant value) are being compared, try to
specify the same data types and attributes. A query that uses the CQE does not use an index
for the named column if the host variable or constant value has a greater precision than the
precision of the column.
If you have different numeric definitions between the column and host variable or constant
value, the queries that use the SQE can use the existing indexes.
If different numeric definitions exist, and no index is available, neither of the two query
engines suggests an index.
To avoid problems for columns and constants being compared, use:
򐂰 The same data type
򐂰 The same scale, if applicable
򐂰 The same precision, if applicable
Figure 9-10 shows an SQL statement that selects all orders with an order total greater than
200. The order_total column is defined as decimal 9.2. In the WHERE clause, we used a
precision of 4. We executed the query twice. On the first one, we specified a logical file. This
way, the query was forced to use CQE. The second one was executed by SQE. Executing this
query with CQE, a table scan is performed. With SQE, an index is found and used.
Figure 9-10 Row selection using different numeric data types in the where clause
242
With SQE, an index can be used if the numeric value is casted to another numeric format,
such as Integer or Float. With CQE, a table scan is performed.
9.2.5 Avoid numeric expressions
Do not use an arithmetic expression as an operand to be compared to a column in a row
selection predicate. The optimizer does not use an index on a column that is being compared
to an arithmetic expression. While this might not cause an index over the column to become
unusable, it prevents any estimates and possibly the use of index scan-key positioning on the
index.
While using CQE, a table scan might be performed. SQE can use the index but must do an
additional relative record scan.
Example 9-6 shows part of an RPG program that selects the orders for next seven days. The
new date is calculated in a numeric expression.
Example 9-6 Using numeric expressions in embedded SQL
D CountOrder
S
5I 0
*---------------------------------------------------------------------------------------C/EXEC SQL
C+ Select Count(Order_Number)
C+ into :CountOrder
C+ from OrdHdr
C+ Where Order_Delivery = Current_Date + 7 Days
C/End-Exec
C
C
CountOrder
Dsply
Return
Instead of a numeric expression in the SQL statement, Example 9-7 shows that a host
variable is defined and the new date is filled from the RPG program.
Example 9-7 Using host variables instead of numeric expressions in SQL
D NextWeek
S
D
D CountOrder
S
5I 0
*----------------------------------------------------------------------------------------C
Eval
NextWeek
= %Date() + %Days(7)
C/EXEC SQL
C+ Select Count(Order_Number)
C+ into :CountOrder
C+ from OrdHdr
C+ Where Order_Delivery = :NextWeek
C/End-Exec
C
C
CountOrder
Dsply
Return
243
Figure 9-11 shows the Visual Explain for both SQL Statements. Notice the table scan when
CQE executes the query. Using a host variable instead, an index can be used.
Figure 9-11 Using numeric expressions or host variables in embedded SQL with CQE
244
When SQE executes the query, in both cases, indexes can be used. But with numeric
expressions, an additional relative record scan is necessary. Figure 9-12 shows the Visual
Explain for both SQL statements when executed by SQE.
Figure 9-12 Using numeric expressions or host variables in embedded SQL with SQE
245
9.2.6 Using the LIKE predicate
The percent sign (%) and the underline (_), when used in the pattern of a LIKE predicate,
specify a character string that is similar to the column value of rows that you want to select.
They can take advantage of indexes when used to denote characters in the middle or at the
end of a character string. However, when used at the beginning of a character string, the %
and _ can prevent DB2 Universal Database for iSeries from using any indexes that might be
defined in the appropriate column to limit the number of rows scanned using index scan-key
positioning.
Figure 9-13 shows the difference when using the LIKE predicate with the % sign in the first
position of the character expression.
Figure 9-13 Using the LIKE predicate in SQL select statements
Note: In V5R3M0, queries that contain the LIKE predicate to select rows are rerouted to
the CQE and cannot benefit from the advantages of the new SQE.
If you search for the first character and want to use the SQE instead of the CQE, you can use
the scalar function substring on the left sign of the equal sign. If you have to search for
additional characters in the string, you can additionally use the scalar function POSSTR. By
splitting the LIKE predicate into several scalar function, you can affect the query optimizer to
use the SQE.
246
Figure 9-14 shows Visual Explain for two select statements that lead to the same result. In the
first case, we use the LIKE predicate. In the second case, we use the scalar functions
SUBSTR() and POSSTR() instead.
Figure 9-14 Replacing the LIKE predicate with SQL scalar functions
247
9.2.7 Avoid scalar functions in the WHERE clause
If a scalar function is used in the WHERE clause, the optimizer might not use an appropriate
existing index. In some cases, you can affect the optimizer by rewriting your query in a
different manner. For example, if you must select all the orders for a specific year or month,
use a range and not the scalar functions YEAR or MONTH.
In the following example, all order totals for the year 2005 are cumulated by month. We first
used the scalar function YEAR to select the desired year in the WHERE clause. In the second
test, we used a date range instead, using the BETWEEN predicate. While in the first example
a table scan is performed, in the second example an index is used.
Figure 9-15 shows the Visual Explain diagrams for both SQL statements.
Figure 9-15 Using ranges instead of scalar functions
9.3 Reorganizing your database
With release V5R3M0, there are several enhancements that can help to analyze and
reorganize the database objects, tables, and indexes for better performance:
򐂰 Index Evaluator
򐂰 Improved reorganization support for tables and physical files
򐂰 Fast Delete support
248
9.3.1 Index Evaluator
Because indexes cannot directly be used in SQL select statements, it is difficult to determine
which one are obsolete. An index is always created with access path maintenance *IMMED.
Therefore, a vast number of indexes can drop down the performance. It is important that you
regularly check the obsolete indexes and remove them.
Index Evaluator is a new function that makes it easier to analyze the indexes (and keyed
logical files) that are helping your SQL and query performance and the indexes that are not
really being used at all. For each index defined over the table, the evaluator returns data
about the last time an index was used in the implementation of a query and provides statistics
about the table data to the query optimizer. In addition, it returns a count for the number of
times the query optimizer used an index object.
In Figure 9-16, you see the output of the Index Evaluator. To initiate it, in iSeries Navigator,
right-click a table object and select Indexes.
Figure 9-16 Index Evaluator
This new capability requires iSeries Access for Windows V5R3 Service Pack 2 on the client
and the PTFs SI12938 and SI12873 installed on the server. The index usage information
does not contain any data about queries that were run before these PTFs are applied.
Therefore, don’t decide on the value of an index before you let the majority of your reports and
queries run.
9.3.2 Improved reorganization support for tables and physical files
In the past, many companies could not use the benefits of reorganized tables. To reorganize
tables, they must be allocated exclusively. This is a problem if the tables are heavily used and
the company works 24/7.
Contrary to the previous releases, you can run the reorganization support online with options
that allow read-level, or even update-level, access to rows in the table while the
reorganization is running.
In V5R3M0, the following new capabilities are added to the reorganize physical file support:
򐂰 Suspended reorganize support
򐂰 Online reorganize
򐂰 Index rebuild options
249
Suspended reorganize support
With V5R3, reorganize support can be suspended and resumed later. This is useful for large
tables that you might not have reorganized in the past because you didn’t have a long enough
window to take the table offline to run the reorganization.
This capability lets you start the reorganize support during a time that suits your business
needs and then suspend the support when that window expires. You effectively get an
incremental reorganization of your table. Then when you have another window available, you
can resume the reorganize support from where it left off last time, or you can start from
scratch if you made significant changes to the table since you suspended the reorganization.
Even this incremental reorganization might increase the efficiency of accessing your table.
Online reorganize
You can also choose whether users can access a table during reorganization (that is, online
reorganize). Online reorganize is beneficial to those who need to reorganize a table but
simply can’t allocate the table exclusively. This gives you two options for allowing user access
to the table while reorganize support is running: allow users read-level access or allow them
read- and update-level access. Depending on your business needs, these two options can
help keep your business running while improving table-access performance.
Keep in mind that reorganization takes longer if you allow users to access the table. Read
access allows you to keep running queries or reports while the table is being reorganized. If
you choose to allow update access, the reorganize support might be unable to put the rows in
the exact key order.
Index rebuilt options
You can choose when (or if) to rebuild indexes built on the table you are reorganizing. This
also has implications on the length of time required for the reorganization. You can choose to
have the indexes maintained during the reorganization (so a rebuild isn’t necessary), after the
reorganization or separate from the reorganization.
Having the indexes maintained during the reorganization provides faster access for query
users because changes to the index are made simultaneously with the data reorganization.
Having the indexes rebuilt at the end provides faster reorganization. However, queries that
typically use the indexes are substantially impaired until the indexes are rebuilt, and any
applications that depend on the indexes attempt to rebuild the indexes during open.
Rebuilding the indexes separate from the reorganization allows the reorganization to be done
faster, but the indexes are not rebuilt as part of the reorganization itself. They are rebuilt in the
background by a separate system process. All of these new capabilities are available on the
Reorganize Physical File Member (RGZPFM) command.
9.3.3 Fast Delete support
As developers move from native I/O to embedded SQL, they often wonder why a Clear
Physical File Member (CLRPFM) command is faster than the SQL equivalent of DELETE
FROM table. The reason is that the SQL DELETE statement deletes a single row at a time. In
i5/OS V5R3, DB2 Universal Database for iSeries has been enhanced with new techniques to
speed up processing when every row in the table is deleted.
If the DELETE statement is not run under commitment control and no WHERE clause is
specified, then DB2 Universal Database for iSeries uses the CLRPFM operation underneath
the covers.
250
If the Delete is performed with commitment control and no WHERE clause is specified, then
DB2 Universal Database for iSeries can use a new method that is faster than the old delete
one row at a time approach. The deletions can still be committed or rolled back, but individual
journal entries for each row are not recorded in the journal. This technique is only used if all
of the following statements are true:
򐂰 The target table is not a view.
򐂰 A significant number of rows is being deleted.
򐂰 The job issuing the DELETE statement does not have an open cursor on the file (not
including pseudo-closed SQL cursors).
򐂰 No other job has a lock on the table.
򐂰 The table does not have an active delete trigger.
򐂰 The table is not the parent in a referential constraint with a CASCADE, SET NULL, or SET
DEFAULT delete rule.
򐂰 The user issuing the DELETE statement has *OBJMGT or *OBJALTER system authority
on the table in addition to the DELETE privilege.
251
252
A
Appendix A.
Tools to check a performance
problem
When a performance problem is occurring, it is important to understand what is happening
with the system, even when you know that you are having a problem with SQL queries. There
are several commands that you can use to see a high level view of what the system is doing.
Use the tools in this appendix to help you in looking at the big picture. They will help you to try
to determine if SQL queries are causing performance problems on the system or if system
tuning needs to be done to prevent SQL query performance problems.
253
WRKACTJOB command
The Work with Active Jobs (WRKACTJOB) command allows you to see what jobs are using
system resources. You can sort on any column. Position the cursor over the column that you
want to examine and press F16 to sort it. You can sort by CPU% to find the job or jobs that
are using most of the CPU.
Figure A-1 shows the Work with Active Jobs display. You can see that the top three jobs are
using the majority of CPU. Refresh this display often to see if the jobs continue to use CPU or
if it was a one-time occurrence. It is important to note the function for the jobs that are using
CPU. You can find the function by looking in the Function column. See if they are building
indexes. If indexes are being built, the function shows IDX-indexname, where indexname is
the name of the index being built.
Work with Active Jobs
CPU %:
99.8
Elapsed time:
2=Change 3=Hold
4=End
8=Work with spooled files
Opt Subsystem/Job
QCLNSYSLOG
QPADEV001V
QPADEV0035
Q1PDR
PRTCPTRPT
CRTPFRDTA
QPADEV004W
QPADEV0018
QRWTSRVR
00:00:04
03/04/05
Active jobs: 1300
5=Work with 6=Release
13=Disconnect ...
User
QPGMR
KLD
TSWEENEY
QPM400
EILEENPI
QSYS
DHUFFMAN
HANS
QUSER
Type
BCH
INT
INT
BCH
BCH
BCH
INT
INT
BCI
CPU %
22.5
18.4
17.5
4.3
3.7
3.7
1.4
1.4
1.3
RCHASCLC
10:13:32
7=Display message
Function
CMD-DLTPRB
CMD-WRKMEDIBRM
CMD-DSPLICKEY
PGM-Q1PBATCH
PGM-QPTBATCH
CMD-CRTPFRDTA
MNU-MAIN
CMD-WRKPRB
Status
RUN
RUN
RUN
RUN
RUN
RUN
RUN
RUN
RUN
More...
Parameters or command
===>
F3=Exit
F5=Refresh
F11=Display elapsed data
F7=Find
F12=Cancel
F10=Restart statistics
F23=More options
F24=More keys
Figure A-1 Work with Active Jobs panel showing jobs using CPU
The Work with Active Jobs display can also show jobs that are using a large amount of I/O
(see Figure A-2). To view the I/O display, you enter the WRKACTJOB command and then
press F11. Then place your cursor in the AuxIO column and press F16 to sort the column.
In the Work with Active Jobs display, the I/O count shown is only reported after an operation
has completed. An example of where the WRKACTJOB I/O count for a job might not match
the Work with System Activity (WRKSYSACT) count is when a blocked fetch is done. The
WRKSYSACT command shows the actual I/O count, where the WRKACTJOB command
does not show the I/O count until the fetch has completed.
In this example, using the WRKACTJOB command, a poor performing SQL statement might
appear as though it is performing little to no I/O, but the WRKSYSACT command shows that
the job is I/O intensive. For more information about the WRKSYSACT command, see
“WRKSYSACT command” on page 255.
It is important to press F11 two more times when looking at the I/O with the WRKACTJOB
command to reach the display that shows the function for the jobs. If the function is
254
IDX-indexname, then the job is building an index. Further investigation must be done to
determine why the job is building an index.
Work with Active Jobs
CPU %:
99.8
Elapsed time:
00:14:25
03/04/05
Active jobs: 1302
RCHASCLC
10:27:54
2=Change 3=Hold
4=End
8=Work with spooled files
Opt Subsystem/Job
PRTSYSRPT
QSCSTT0001
QSCSTT0004
QPADEV0048
QPADEV004W
QPADEV003N
QRWTSRVR
QPADEV0018
AMQPCSEA
Type
BCH
BCH
BCH
INT
INT
INT
BCI
INT
BCH
===>
F3=Exit
F5=Refresh
F11=Display thread data
5=Work with 6=Release
7=Display message
13=Disconnect ...
--------Elapsed--------Pool Pty
CPU Int
Rsp AuxIO CPU %
2 50
62.3
29230
1.5
2 25
5743.6
22403
.3
2 25
6025.5
22346
.3
3 20
4.0 109
2.7 15138
.0
3 20
36282.0
0
.0
6551
.9
3 20
1.5
37
2.8
3733
.0
2 20
33036.7
3601
.5
3 20
2747.9
0
.0
3389
.4
2 35
12.0
2434
.0
More...
F7=Find
F12=Cancel
F10=Restart statistics
F23=More options
F24=More keys
Figure A-2 Work with Active Jobs panel showing I/O used
WRKSYSACT command
WRKSYSACT command is provided with the Performance Tools. This command is helpful in
finding jobs that use system resources. The advantage of using this command over the
WRKACTJOB command is that the WRKSYSACT command shows the Licensed Internal
Code (LIC) tasks in addition to the active jobs in the system.
The WRKSYSACT command, by default, sorts on CPU utilization as shows in Figure A-3. It is
important to note the elapsed time. To watch for jobs using CPU, press F10 often to see if the
same jobs stay near the top of the list of jobs using CPU. These are the jobs that you want to
determine what they are doing.
One way to determine what a job is doing is to look at the function of the job using the
WRKACTJOB command as shown in Figure A-1. You can also use the Work with Jobs
(WRKJOB) command to see what the job is doing. For more information about the WRKJOB
command, see “WRKJOB command” on page 259. If it is known that the jobs using CPU are
also using SQL, then you want to look at Database Monitor data or use other tools to try to
capture the performance problem.
Appendix A. Tools to check a performance problem
255
Work with System Activity
Automatic refresh in
Elapsed time . . . .
Number of CPUs . . .
Overall DB CPU util
seconds . . . . . . . . . . . . . . .
. . :
00:00:02
Average CPU util .
. . . :
4
Maximum CPU util .
. . . :
7.7
Minimum CPU util .
Current processing
1=Monitor job
5=Work with job
RCHASCLC
03/04/05 10:24:50
. . . .
5
. . . :
99.9
. . . . : 101.8
. . . . :
98.8
capacity:
3.00
Job or
Opt Task
QCLNSYSLOG
QPADEV0035
QPADEV001V
Q1PDR
CRTPFRDTA
PRTCPTRPT
QPADEV0015
QPADEV004W
User
QPGMR
TSWEENEY
KLD
QPM400
QSYS
EILEENPI
PEGGYCL
DHUFFMAN
F3=Exit
F10=Update list
F24=More keys
Number
640427
639636
640239
640481
640539
641086
641174
640056
Thread
00000088
00000002
0000008A
00000416
00000029
00000022
00000016
00000040
F11=View 2
Pty
10
23
20
50
50
50
1
23
Total DB
Async CPU
I/O Util
0
.0
0
.0
0
.0
0 7.1
0
.0
0
.0
0
.0
0
.5
More...
F19=Automatic refresh
CPU
Util
24.3
22.4
21.7
4.9
4.7
4.5
1.7
1.5
F12=Cancel
Total
Sync
I/O
0
0
0
0
0
0
4
28
Figure A-3 Work with System Activity panel sorted by CPU
The WRKSYSACT command can also sort on different resources, such as I/O. You can
resequence the list by selecting F16. Then you see the Select Type of Sequence display (see
Figure A-4).
Select Type of Sequence
Type option, press Enter.
Option
. . . .
2 1.
2.
3.
4.
5.
6.
7.
Sequence
Sequence
Sequence
Sequence
Sequence
Sequence
Sequence
by
by
by
by
by
by
by
CPU
I/O
net storage
allocated storage
deallocated storage
database CPU
total waiting time
F12=Cancel
Figure A-4 Sequence options for the WRKSYSACT command
In this example, we type option 2 to sort by I/O. Then you see the Work with System Activity
display shown in Figure A-5. It is important to refresh the display with F10 often to see if the
same jobs are doing a lot of I/O. It is also good to notice if the I/O is synchronous or
asynchronous. In most cases, asynchronous I/O is more desirable. Asynchronous disk I/O
means the job can do other work while waiting for disk I/O to complete. Synchronous disk I/O
is when a job has to wait for disk I/O work to be done before doing other work.
When you know the job that is using I/O, then determine what the job is doing. You can use
the WRKACTJOB command to find the job and see what function the job is in at the time as
256
shown in Figure A-1 on page 254. You can also use the WRKJOB command to see what the
job is doing. For more information about the WRKJOB command, see “WRKJOB command”
on page 259.
If SQL is being done, then Database Monitor data can determine what is being done. For
more information about gathering Database Monitor data, see Chapter 4, “Gathering
database SQL performance data” on page 51.
Work with System Activity
Automatic refresh in
Elapsed time . . . .
Number of CPUs . . .
Overall DB CPU util
seconds . . . . . . . . . . . . . . .
. . :
00:00:02
Average CPU util .
. . . :
4
Maximum CPU util .
. . . :
7.7
Minimum CPU util .
Current processing
1=Monitor job
5=Work with job
RCHASCLC
03/04/05 10:24:50
. . . .
5
. . . :
99.9
. . . . : 101.8
. . . . :
98.8
capacity:
3.00
Job or
Opt Task
SMPO0001
GLIDDEN51E
LTWAS51ND
PRTSYSRPT
QYPSJSVR
GNETSIZ
NODEAGENT
MICK51EXP
User
Number
Thread
QEJBSVR
QEJBSVR
EILEENPI
QYPSJSVR
AJMAC
QEJBSVR
QEJBSVR
620898
632143
641085
624351
639853
621978
636665
0000000C
00002D98
00000089
0000002C
00000027
00000015
000000D2
F14=Display jobs only
F24=More keys
F15=Display tasks only
Pty
0
0
26
50
0
50
0
0
CPU
Util
1.0
.3
.4
1.3
.2
1.1
.2
.2
Total
Sync
I/O
0
173
144
46
131
3
125
115
Total DB
Async CPU
I/O Util
1786
.0
0
.0
0
.0
91
.0
0
.0
124
.0
0
.0
0
.0
More...
F16=Resequence
Figure A-5 Work with System Activity panel sorted by I/O
WRKSYSSTS command
The Work with System Status (WRKSYSSTS) command provides a global view of the
system. You can get a better view if you press F21 and select the advanced assistance level.
You can press F10 to restart the statistics. It is best to look at the data after a couple of
minutes have elapsed. F5 refreshes the display until you see 00:02:00.
Figure A-6 shows a Work with System Status display. When you have a performance
problem, check the Work with System Status display to see if your % CPU used is higher than
you normally run. You can also look at the pools to see if you have abnormally high faulting or
paging.
What do you do when % CPU used seems high? One item to check is the % DB Capability. If
the % DB Capability is also high, then it is an indication that there SQL activity is occurring.
Use the WRKSYSACT command to display the jobs using CPU. See “WRKSYSACT
command” on page 255 for more information.
257
򐂰 If a high priority job (low number) is using a lot of CPU, greater than 50% for an extended
period of time, then the job can cause the entire system to have poor response times. If it
is found that one or a few jobs are using the majority of CPU, then ask:
– Is the priority of the job really appropriate?
– Is the job running in the correct environment? For example, if the job is interactive, is it
be better suited running in batch?
– What is the job doing?
򐂰 If the CPU utilization is high, greater than 80%, and all jobs seem to have an equal but
small CPU percent, this can mean that there are too many active jobs on the system.
Work with System Status
% CPU used . . .
% DB capability
Elapsed time . .
Jobs in system .
% perm addresses
% temp addresses
Sys
Pool
1
2
3
4
5
6
7
8
9
Pool
Size M
788.06
2754.35
4785.78
44.91
94.20
1.25
94.20
94.20
763.33
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Reserved
Size M
227.82
2.17
.00
.00
.00
.00
.00
.00
.17
:
:
:
:
:
:
2.7
.0
00:02:00
14182
.202
.910
03/04/05
System ASP . . . . . . . :
% system ASP used . . . :
Total aux stg . . . . . :
Current unprotect used . :
Maximum unprotect . . . :
Max ----DB----Act Fault Pages
+++++
.0
.0
9514
.0
.8
233
.0
.0
11
.0
.0
5
.0
.0
6
.0
.0
24
.0
.0
5
.0
.0
23
.0
.0
--Non-DB--Fault Pages
.9 1.3
11.4 42.8
4.1 4.8
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
ActWait
18.3
2148
32.8
.0
.0
.0
.0
.0
83.5
RCHASCLC
13:52:55
1922 G
88.2414
2055 G
6944 M
7476 M
Wait- ActInel Inel
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
.0
Bottom
===>
F21=Select assistance level
Figure A-6 Work with System Status panel
For more information about the WRKSYSSTS command and how to view the data via iSeries
Navigator, see Managing OS/400 with Operations Navigator V5R1 Volume 5: Performance
Management, SG24-6565.
WRKOBJLCK command
Using the Work with Object Lock (WRKOBJLCK) command on user profiles can help narrow
down clients that are having performance problems. For example, a user is using an ODBC
connection and is complaining about having a performance problem. To find the job that the
user is running on the iSeries, enter the following command:
WRKOBJLCK OBJ(QSYS/userprofile) OBJTYPE(*USRPRF)
In this example, userprofile refers to the user’s iSeries user ID.
A panel is displayed that shows a list of jobs that the user profile has locked. You can work
with each job to see if you can isolate the one that is having the problem. You can look at the
call stack and objects locked to see a picture of what the job is doing.
If any program beginning with QSQ is found in the call stack or QDBGETMQO is found, then
SQL is being used. Refer to Chapter 3, “Overview of tools to analyze database performance”
258
on page 31, to learn about other tools that you can use to further analyze the job, after the
correct job is found.
WRKJOB command
You can use the Work with Job (WRKJOB) command to determine what a job is doing. It
includes options to look at the job. Some of the following options might be helpful to check for
jobs that have performance problems.
򐂰 Use option 1 to find the pool in which the job is running. The subsystem and subsystem
pools are shown:
An example is:
Subsystem . . . . . . . . . . . . . . . . . :
Subsystem pool ID . . . . . . . . . . . . :
QINTER
2
To find the system pool in which the job is running, enter the Work with Subsystems
(WRKSBS) command.
Figure A-7 shows the Work with Subsystems display for our example. As you can see, the
subsystem pool ID is 2 and the subsystem is QINTER. We go to subsystem QINTER and
look under the subsystem pool column of 2, which shows the job is using system pool 3.
Work with Subsystems
System:
RCHASCLC
4=End subsystem
5=Display subsystem description
8=Work with subsystem jobs
Opt Subsystem
QASE5
QASE51
QBATCH
QCMN
QCTL
QEJBAS51
QHTTPSVR
QINTER
QMQM
QSERVER
Total
-----------Subsystem Pools-----------Storage (M) 1
2
3 4
5 6
7
8 9 10
.00 2
8
.00 2
9
1.25 2
6
.00 2
.00 2
.00 2
.00 2
.00 2
3
4
5
.00 2
.00 2
More...
===>
F3=Exit
F5=Refresh F11=Display system data
F14=Work with system status
F12=Cancel
Figure A-7 Work with Subsystems panel
򐂰 Use option 3 to give the job run attributes, if the job is active. This option is helpful in
determining how much CPU and temporary storage are being used in the job. High
temporary storage use can be the result of temporary indexes being built.
򐂰 Use option 4 to see all the spooled files for the job. Check this to see if any unexpected
spooled files are being created.
򐂰 Use option 10 to see the job log for the job. It is important to review the job log. Chapter 3,
“Overview of tools to analyze database performance” on page 31, explains how to turn on
debug messages to capture optimizer messages when running SQL.
259
򐂰 Use option 11 to view the call stack for the job. The example in Figure A-8 shows
QDBGETMQO at the bottom of the call stack. QDBGETMQO is when the SQL Query
Engine (SQE) is getting rows. SQE is discussed in Chapter 2, “DB2 Universal Database
for iSeries performance basics” on page 9.
If QDBGETM is displayed at the bottom of the call stack, either the Classic Query Engine
(CQE) is fetching rows or there is native I/O such as in RPG. It is important to note any
user programs that are at the bottom of the call stack. If a user program is found, use the
Print SQL Information (PRTSQLINF) command to see if the program contains SQL. For
details about PRTSQLINF, see Chapter 3, “Overview of tools to analyze database
performance” on page 31.
Use F10 to monitor whether the programs in the call stack change. If any program
beginning with QSQ is found in the call stack or QDBGETMQO is found, then SQL is being
used.
Display Call Stack
Job:
Thread:
QPADEV0027
Rqs
Lvl
User:
PEGGYCL
Number:
System:
647778
RCHASCLC
00000046
Program
or
Procedure
QSQIMAIN
QSQISE
QQUDA
QQURA
QQURB
QDBGETMQO
F3=Exit
F16=Job menu
Library
QSQL
QSQL
QSYS
QSYS
QSYS
QSYS
Statement
Instruction
05CA
0707
03CD
0087
0677
0000002573
F10=Update stack F11=Display activation group
F17=Top F18=Bottom
F22=Display entire name
Bottom
F12=Cancel
Figure A-8 Display Call Stack panel, when using option 11 from the WRKJOB command
iDoctor for iSeries Job Watcher
For a more in depth analysis of a performance problem, you can use the advanced analysis
tool called iDoctor for iSeries Job Watcher, which we refer to as Job Watcher. Job Watcher is
made up of two parts:
򐂰
򐂰
Tools for collecting data
Tools for analyzing and viewing the collected data
A typical situation for deciding to use the Job Watcher is for a job that is taking a long time to
run but is hardly using any CPU resource and disk I/Os are not particularly excessive. Job
Watcher is an excellent tool to help you determine job waits, seizes, and other types of
contention. Identifying why a job or multiple jobs or threads are “not doing anything when they
should be,” is a primary situation to demonstrate a key set of Job Watcher capabilities.
Job Watcher returns near real-time information about a selected set of jobs, threads, LIC
tasks, or all three. It is similar in sampling function to the WRKACTJOB and WRKSYSACT
system commands, where each refresh computes delta information for the ending snapshot
interval. In Job Watcher, these refreshes can be set to occur automatically, even as frequently
260
as every 5 seconds. Better yet, Job Watcher harvests the data from the jobs, threads, or tasks
being watched in a manner that does not impact other jobs on the system while it is collecting.
Job Watcher collected data includes the following information among other details:
򐂰 Standard WRKSYSACT type information
–
–
–
–
CPU
DASD I/O breakdown
DASD space consumption
For jobs or threads, the user profile under which the job or thread is running
For prestart server jobs that were started under user profile QUSER, you can see the
user profile that is currently being serviced by that job/thread, rather than QUSER.
򐂰 Expanded details on types of waits and object lock or seize conditions
򐂰 Last run SQL statements syntax
򐂰 Program or procedure call stack, 1000 levels deep
You can download Job Watcher from the following Web site:
http://www.ibm.com/eserver/iseries/support/i_dir/idoctor.nsf
Select downloads from the left pane and then select the appropriate download option. You
can sign up for a 45-day trial to use the product.
For further information about using Job Watcher, see the IBM iDoctor iSeries Job Watcher:
Advanced Performance Tool, SG24-6474.
261
262
Related publications
The publications listed in this section are considered particularly suitable for a more detailed
discussion of the topics covered in this redbook.
IBM Redbooks
For information about ordering these publications, see “How to get IBM Redbooks” on
page 264. Note that some of the documents referenced here may be available in softcopy
only.
򐂰 Advanced Functions and Administration on DB2 Universal Database for iSeries,
SG24-4249
򐂰 IBM iDoctor iSeries Job Watcher: Advanced Performance Tool, SG24-6474
򐂰 Stored Procedures, Triggers and User Defined Functions on DB2 Universal Database for
iSeries, SG24-6503
򐂰 Managing OS/400 with Operations Navigator V5R1 Volume 5: Performance Management,
SG24-6565
򐂰 Preparing for and Tuning the V5R2 SQL Query Engine on DB2 Universal Database for
iSeries, SG24-6598
򐂰 Using AS/400 Database Monitor and Visual Explain To Identify and Tune SQL Queries,
REDP-0502
Other publications
These publications are also relevant as further information sources:
򐂰 DB2 Universal Database for iSeries Database Performance and Query Optimization
http://publib.boulder.ibm.com/infocenter/iseries/v5r3/index.jsp?topic=/rzajq/
rzajqmst.htm
򐂰 Star Schema Join Support within DB2 UDB for iSeries - Version 3
http://www-1.ibm.com/servers/enable/site/education/abstracts/16fa_abs.html
Online resources
These Web sites are also relevant as further information sources:
򐂰 Information Center
򐂰 iDoctor for iSeries Job Watcher
http://www.ibm.com/eserver/iseries/support/i_dir/idoctor.nsf
263
How to get IBM Redbooks
You can search for, view, or download Redbooks, Redpapers, Hints and Tips, draft
publications and Additional materials, as well as order hardcopy Redbooks or CD-ROMs, at
this Web site:
ibm.com/redbooks
Help from IBM
IBM Support and downloads
ibm.com/support
IBM Global Services
ibm.com/services
264
Index
Symbols
% CPU used 257
% DB Capability 257
*HEX 27
Numerics
1000 Record 77
1000 Record (SQL statement summary) 80
3000 Record 77, 82
3001 Record 77, 83
3002 Record 77
3002 Record (temporary index created) 85
3003 Record 77, 86
3004 Record 77, 87
3005 Record 78
3006 Record 77, 87
3007 Record 78, 88
3008 Record 78
3010 Record 78, 89
3014 Record 78, 90
3015 Record 90
3018 Record 78
3019 Record 78, 90
3021 Record 78
3022 Record 78
3023 Record 78
3025 Record 78
3026 Record 78
3027 Record 78
3028 Record 78
3029 Record 78
3030 Record 78
5002 Record 78
5722PT1 180
A
access methods 230
Index probe 230
Index scan 230
Table probe 230
Table scan 230
access plan 22
rebuilt 159
Access Plan Rebuild Information report 110
Alternate Collating Sequence 19
analysis tools for database performance 31
ANDing 11
API support for Memory Resident Database Monitor 47
asynchronous disk I/O 256
authentication 5
automatic summary table 175
B
Basic Statement Information report 105
binary-radix tree index 10
bitmap index 11
bitmap indexing 11
Bitmap Information report 118
Boolean arithmetic 11
C
CFGPFRCOL (Configure Performance Collection) command 182
Change Query Attributes (CHGQRYA) CL command
QRYTIMLMT 39
CHGQRYA (Change Query Attributes) CL command 39
CHGSYSVAL QQRYTIMLMT 39
Classic Query Engine (CQE) 9, 14, 260
Statistics Manager 19
temporary index 85
Collection Services 180
data to identify jobs using system resources 179
start 180
Collector APIs 182
communications 5
Component Report 180
compression algorithm 11
Configure Performance Collection (CFGPFRCOL) CL
command 182
connection keyword
ODBC 56
OLE DB 58
connection properties, OLE DB 57
correlated subquery 222
CPU bound 212
CQE (Classic Query Engine) 9, 14, 260
CREATE ENCODED VECTOR INDEX statement 11, 13
Create Logical File (CRTLF) command 19, 151
Create Performance Data (CRTPFRDTA) command 183
CRTLF (Create Logical File) command 19, 151
CRTPFRDTA (Create Performance Data) command 183
Current SQL for a Job function 31–32
D
Data Access Primitives 16
Data Conversion Information report 117
data conversion problems 65
Data Definition Language (DDL) 13
Data Manipulation Language (DML) 13
data processing 5
data source name, ODBC 54
database architecture prior to V5R2 15
Database Monitor 15, 180
data organization in a table 80
end 54
265
exit program 59
global data fields 78
import into iSeries Navigator 69
JDBC client 58
ODBC clients 54
OLE DB client 57
query examples 137
record types 77
start 52
tips to analyze files 135
Database Monitor record types
1000 Record 77, 80
3000 Record 77, 82
3001 Record 77, 83
3002 Record 77, 85
3003 Record 77, 86
3004 Record 77, 87
3005 Record 78
3006 Record 77
3007 Record 78, 88
3008 Record 78
3010 Record 78, 89
3014 Record 78, 90
3015 Record 90
3018 Record 78
3019 Record 78, 90
3021 Record 78
3022 Record 78
3023 Record 78
3025 Record 78
3026 Record 78
3027 Record 78
3028 Record 78
3029 Record 78
3030 Record 78
5002 Record 78
Database Monitor table
additional index 136
SLQ view 136
subset for faster analysis 135
database performance analysis tools 31
Database Performance Monitors 43
database reorganization 248
DDL (Data Definition Language) 13
debug information messages 15
debug messages 31, 36
detail row 53
Detailed Database Monitor 52, 75
end 54
start 52, 65
Detailed Monitor 31, 44
dimension table 25
disk I/O counts 193
display performance data 187
distinct key list 11
Distinct Processing Information report 117
DML (Data Manipulation Language) 13
dynamic bitmap 11
dynamic SQL 22
EXECUTE IMMEDIATE statement 22
266
SQL PREPARE 22
E
encoded-vector index (EVI) 10–11
recommended use 12
End Database Monitor (ENDDBMON) command 44, 54
ENDDBMON (End Database Monitor) command 44, 54
equi-join predicate 25–26
Error Information report 119
EVI (encoded-vector index) 10–11
recommended use 12
exit program 59
expert cache 160
Explain Only 200
Explain SQL 34
explainable statement 214
Extended Detailed report 105
extended dynamic SQL 22
external table description 63
Memory Resident Database Monitor 48
F
fact table 25
Fast Delete support 250
full open 109
analysis 142
G
global field
QQ19 79
QQI5 78
QQJFLD 78
QQJNUM 78
QQJOB 78
QQRID 78
QQTIME 78
QQUCNT 78
QQUSER 78
QVC102 79
good working
condition 7
situation 7
Governor Timeout Information report 117
Group By Information report 116
H
hash grouping 112
hash join 112
hash key 225
Hash Table Information report 111
I
I/O bound 212
iDoctor for iSeries Job Watcher 260
index
advised 37, 155
over the Database Monitor table 136
temporary 151
Index Advised Information report 111
Index Advisor 31, 39, 204, 235
advise for encoded-vector index 40
radix index suggestion 39
Index Create Information report 113
Index Evaluator 31, 40, 249
index only access (IOA) 232
index optimization data 78
index rebuilt options 250
index scan-key selection 153
Index Used Information report 112
indexing 8
strategy 230
tips 232
IOA (index only access) 232
iSeries Navigator
Create Database Files Now 184
Current SQL for a Job 32
Graph History 183, 189
import of Database Monitors 69
monitor data collection 59
SQL Performance Monitors 75
Visual Explain 49
isolation level 147
QAQQ3007 48
QAQQ3008 48
QAQQ3010 48
QAQQQRYI 48
QAQQTEXT 48
QQQCSDBM 47
QQQDSDBM 47
QQQESDBM 47
QQQQSDBM 47
QQQSSDBM 47
external table description 48
Memory-based Database Performance Monitor 44
MI (machine interface) 15
monitor data collection 52
iSeries Navigator 59
MQT (materialized query table) 175
MQT record types 92
J
O
JDBC client, enabling Database Monitor 58
JOB parameter 45
Job Watcher 260
jobs using system resources 179
K
key value 10
L
leaf node 10
LIKE predicate 246
List Explainable Statements 214, 218
Lock Escalation Information report 118
logical file, avoid use in select statements 238
Lookahead Predicate Generation (LPG) 9, 27
LPG (Lookahead Predicate Generation) 9, 27
M
machine interface (MI) 15
Management Central System Monitors 183, 191
management collection object (*MGTCOL) 181
materialized query table (MQT) 175
materialized view 175
Memory Resident Database Monitor 46, 52
analysis of data conversion problems 65
API support 47
QAQQ3000 48
QAQQ3001 48
QAQQ3002 48
QAQQ3003 48
QAQQ3004 48
N
nonreusable ODP mode 80
nonsensical query 17
normalized tables 25
numeric data type conversion 242
numeric expression 243
object-oriented design 15, 17
ODBC 17
connection keywords 56
data source name 54
Database Monitor 54
ODP (open data path) 215
OLE DB
client 57
connection keywords 58
connection properties 57
OLTP (online transaction processing) 10
online reorganize 250
online transaction processing (OLTP) 10
open data path (ODP) 37, 215
reusable 44, 147, 215
Open Information report 109
open processing time 6–7
OPNQRYF command 219
optimization record 78
optimization time 6–7
Optimizer Information report 106
Optimizer Timeout Information report 117
ORing 11
P
perfect index 10
PERFORM menu 180
Performance Management APIs 180
Performance Tools 255
persistent indexes 10
Plan Cache 16
definition 24
Index
267
predefined query report 119
Predictive Query Governor 39
primary key column 153
Print performance report 184
print SQL information 31, 34
Print SQL Information (PRTSQLINF) command 34, 44,
260
proactive tuning 123
Procedure Call Information report 117
PRTSQLINF (Print SQL Information) command 34, 44,
260
pseudo open 109, 142, 145
Q
QAQQINI option
FORCE_JOIN_ORDER 26
IGNORE_DERIVED_INDEX 19
STAR_JOIN 26
QAQQINI parameter
MESSAGES_DEBUG 37
QUERY_TIME_LIMIT 39
QDBFSTCCOL system value 167
QDBGETM 260
QDBGETMQO 260
QPFRADJ system value 5
QQRID value 77
QRWTSRVR jobs 6
QSQPRCED API 22
QSQSRVR 6
query analysis 134
query attributes and values 200
Query Dispatcher 16
query engine 13
query feedback 15
Query Implementation Graph 200
query report, predefined 119
query sorting 163
QUSRMBRD API 41
QZDASOINIT jobs 5
R
radix index 10
suggestion 39
reactive tuning 123
record types in Database Monitor 77
Redbooks Web site 264
Contact us xi
REFRESH TABLE 175
relative record number (RRN) 12, 231
avoid using to access data 241
RENAME 19
reorganization support for tables and physical files 249
report
Access Plan Rebuild Information 110
Basic Statement Information 105
Bitmap Information 118
Distinct Processing Information 117
Error Information 119
Governor Timeout Information 117
268
Group By Information 116
Hash Table Information 111
Index Advised Information 111
Index Create Information 113
Index Used Information report 112
Lock Escalation Information 118
Open Information 109
Optimizer Information 106
Optimizer Timeout Information 117
Procedure Call Information 117
query tuning example 123
Row Access Information 118
Sort Information 116
Start and End Monitor Information 119
Subquery Information 118
Table Scan Information 107
Temporary File Information 115
Union Merge Information 118
reusable ODP 44, 147, 215
mode 80
Row Access Information report 118
RRN (relative record number) 12, 231
run time 6
S
scalar function, avoid in WHERE clause 248
secondary key column 153
SELECT *, avoid use in select statements 240
select statement
avoid use of logical files 238
avoid use of SELECT * 240
SELECT/OMIT DDS keyword 19
select/omit logical file 151
snowflake schema or model 25
Sort Information report 116
sparse index 113, 151
SQE (SQL Query Engine) 9, 14, 260
SQE Optimizer 16
SQL
analysis of operation types 141
elapsed time 140
embedded 6
performance report information 105
problem causing requests 138
total time spent in 139
SQL (Standard Query Language) 13
SQL ALIAS 135
SQL package 23, 34
advantages 23
deletion 23
SQL Performance Monitor 43
considerations for iSeries Navigator 75
Detailed Monitor 44
Memory Resident Database Monitor 46
properties 71
query tuning example 123
Summary Monitor 46
types 52
Visual Explain 49
SQL Query Engine (SQE) 9, 14, 260
Data Access Primitives 22
node-based implementation 17
staged implementation 18
statistics advised 167
SQL Script Center 199
SQL statement optimization 238
SQL statement summary (1000 Record) 80
SQL view for Database Monitor table 136
SQLCODE 76
SQLSTATE 76
Standard Query Language (SQL) 13
Star Join Schema 9, 25
query 25–26
restrictions and considerations 27
star schema 25
Start and End Monitor Information report 119
Start Database Monitor (STRDBMON) command 44, 52
JOB parameter 45
OUTFILE parameter 45
OUTMBR parameter 45
Start Debug (STRDBG) command 36, 38, 44
Start Performance Collection (STRPFRCOL) CL command 180, 182
Start Server Job (STRSRVJOB) CL command 38
static SQL 22
statistics 8
advised 167
cardinality of values 21
frequent values 21
metadata information 21
selectivity 21
Statistics and Index Advisor 204
Statistics Manager 15–16, 167
STRDBG (Start Debug) command 36, 38, 44
STRDBMON (Start Database Monitor) command 44, 52
STRPFRCOL (Start Performance Collection) CL command 180, 182
STRSRVJOB (Start Server Job) CL command 38
Subquery Information report 118
subsystem pools 259
Summary Database Monitor 52
when to use 64
Summary Monitor 31, 44, 46, 75
Summary Reports
Detailed Performance Monitor 103
Memory-Resident 102
suspended reorganize support 250
symbol table 11
symmetric multiprocessing 18
synchronous disk I/O 256
system pool 259
system resources used by jobs 179
U
Union Merge Information report 118
unique count 106
user display I/O 5
V
vector 12
very large database (VLDB) 11
Visual Explain 15, 31, 48
attributes and values 211
Explain Only 200
icons 220
Index Advisor 235
iSeries Navigator 49
navigating 200
non-SQL interface 219
query environment 211
Run and Explain 200
SQL Performance Monitor 49
toolbar 202
what is 198
VLDB (very large database) 11
W
WHERE clause, avoidance of scalar functions 248
Work with Active Jobs (WRKACTJOB) command 254
Work with Jobs (WRKJOB) command 255
Work with Object Lock (WRKOBJLCK) command 258
Work with Subsystems (WRKSBS) command 259
Work with System Activity (WRKSYSACT) command
254–255
Work with System Status (WRKSYSSTS) command 257
WRKACTJOB (Work with Active Jobs) command 254
WRKJOB (Work with Jobs) command 255
WRKOBJLCK (Work with Object Lock) command 258
WRKSBS (Work with Subsystems) command 259
WRKSYSACT (Work with System Activity) command
254–255
WRSYSSTS (Work with System Status) command 257
T
table scan 148
Table Scan Information report 107
Temporary File Information report 115
temporary index analysis 151
temporary index created (3003 Record) 85
temporary result 37
Index
269
270
SQL Performance Diagnosis on IBM
DB2 Universal Database for iSeries
(0.5” spine)
0.475”<->0.873”
250 <-> 459 pages
Back cover
®
SQL Performance
Diagnosis
on IBM DB2 Universal Database for iSeries
Discover the tools to
identify SQL
performance
problems
Unleash the
capabilities of the SQL
Performance
Monitors
Learn to query the
Database Monitor
performance data
The goal of database performance tuning is to minimize the
response time of your queries. It is also to optimize your server’s
resources by minimizing network traffic, disk I/O, and CPU time.
This IBM Redbook helps you to understand the basics of identifying
and tuning the performance of Structured Query Language (SQL)
statements using IBM DB2 Universal Database for iSeries. DB2
Universal Database for iSeries provides a comprehensive set of
tools that help technical analysts tune SQL queries. The SQL
Performance Monitors are part of the set of tools that IBM i5/OS
provides for assisting in SQL performance analysis since Version 3
Release 6. These monitors help to analyze database performance
problems after SQL requests are run.
This redbook also presents tips and techniques based on the SQL
Performance Monitors and other tools, such as Visual Explain. You’ll
find this guidance helpful in gaining the most out of both DB2
Universal Database for iSeries and query optimizer when using SQL.
INTERNATIONAL
TECHNICAL
SUPPORT
ORGANIZATION
BUILDING TECHNICAL
INFORMATION BASED ON
PRACTICAL EXPERIENCE
IBM Redbooks are developed
by the IBM International
Technical Support
Organization. Experts from
IBM, Customers and Partners
from around the world create
timely technical information
based on realistic scenarios.
Specific recommendations
are provided to help you
implement IT solutions more
effectively in your
environment.
For more information:
ibm.com/redbooks
SG24-6654-00
ISBN 0738497487

Front cover SQL Performance Diagnosis on IBM DB2 Universal

Transcription

Similar documents

Advanced Webi Techniques from the Jedi Academy

Contact no.(M):9780893349 jalandhar city

Five Techniques for Loading Flat File Data into Oracle Tables

2016 SQL stakeholder meeting - summary

The maximum number of users permitted by your

SQL Basics for RPG Developers

A Scalable and Resilient Platform for Microsoft SQL

Webgate400 datasheet - Webgate400

document - MWC brokerage event